]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | This is a rough history of garbage collector bugs and versions. |
2 | ||
3 | This has been maintained with varying diligence over the years. | |
4 | ||
5 | I made an attempt to include recent contributors here. I apologize for any | |
6 | omissions. | |
7 | ||
8 | RECENT UPDATES ARE RECORDED IN ../ChangeLog FILE INSTEAD TO CONFORM TO | |
9 | MORE COMMON CONVENTIONS. | |
10 | ||
11 | ------------------------- | |
12 | ||
13 | Version 1.3 and immediately preceding versions contained spurious | |
14 | assembly language assignments to TMP_SP. Only the assignment in the PC/RT | |
15 | code is necessary. On other machines, with certain compiler options, | |
16 | the assignments can lead to an unsaved register being overwritten. | |
17 | Known to cause problems under SunOS 3.5 WITHOUT the -O option. (With | |
18 | -O the compiler recognizes it as dead code. It probably shouldn't, | |
19 | but that's another story.) | |
20 | ||
21 | Version 1.4 and earlier versions used compile time determined values | |
22 | for the stack base. This no longer works on Sun 3s, since Sun 3/80s use | |
23 | a different stack base. We now use a straightforward heuristic on all | |
24 | machines on which it is known to work (incl. Sun 3s) and compile-time | |
25 | determined values for the rest. There should really be library calls | |
26 | to determine such values. | |
27 | ||
28 | Version 1.5 and earlier did not ensure 8 byte alignment for objects | |
29 | allocated on a sparc based machine. | |
30 | ||
31 | Version 1.8 added ULTRIX support in gc_private.h. | |
32 | ||
33 | Version 1.9 fixed a major bug in gc_realloc. | |
34 | ||
35 | Version 2.0 introduced a consistent naming convention for collector | |
36 | routines and added support for registering dynamic library data segments | |
37 | in the standard mark_roots.c. Most of the data structures were revamped. | |
38 | The treatment of interior pointers was completely changed. Finalization | |
39 | was added. Support for locking was added. Object kinds were added. | |
40 | We added a black listing facility to avoid allocating at addresses known | |
41 | to occur as integers somewhere in the address space. Much of this | |
42 | was accomplished by adapting ideas and code from the PCR collector. | |
43 | The test program was changed and expanded. | |
44 | ||
45 | Version 2.1 was the first stable version since 1.9, and added support | |
46 | for PPCR. | |
47 | ||
48 | Version 2.2 added debugging allocation, and fixed various bugs. Among them: | |
49 | - GC_realloc could fail to extend the size of the object for certain large object sizes. | |
50 | - A blatant subscript range error in GC_printf, which unfortunately | |
51 | wasn't exercised on machines with sufficient stack alignment constraints. | |
52 | - GC_register_displacement did the wrong thing if it was called after | |
53 | any allocation had taken place. | |
54 | - The leak finding code would eventually break after 2048 byte | |
55 | byte objects leaked. | |
56 | - interface.c didn't compile. | |
57 | - The heap size remained much too small for large stacks. | |
58 | - The stack clearing code behaved badly for large stacks, and perhaps | |
59 | on HP/PA machines. | |
60 | ||
61 | Version 2.3 added ALL_INTERIOR_POINTERS and fixed the following bugs: | |
62 | - Missing declaration of etext in the A/UX version. | |
63 | - Some PCR root-finding problems. | |
64 | - Blacklisting was not 100% effective, because the plausible future | |
65 | heap bounds were being miscalculated. | |
66 | - GC_realloc didn't handle out-of-memory correctly. | |
67 | - GC_base could return a nonzero value for addresses inside free blocks. | |
68 | - test.c wasn't really thread safe, and could erroneously report failure | |
69 | in a multithreaded environment. (The locking primitives need to be | |
70 | replaced for other threads packages.) | |
71 | - GC_CONS was thoroughly broken. | |
72 | - On a SPARC with dynamic linking, signals stayed diabled while the | |
73 | client code was running. | |
74 | (Thanks to Manuel Serrano at INRIA for reporting the last two.) | |
75 | ||
76 | Version 2.4 added GC_free_space_divisor as a tuning knob, added | |
77 | support for OS/2 and linux, and fixed the following bugs: | |
78 | - On machines with unaligned pointers (e.g. Sun 3), every 128th word could | |
79 | fail to be considered for marking. | |
80 | - Dynamic_load.c erroneously added 4 bytes to the length of the data and | |
81 | bss sections of the dynamic library. This could result in a bad memory | |
82 | reference if the actual length was a multiple of a page. (Observed on | |
83 | Sun 3. Can probably also happen on a Sun 4.) | |
84 | (Thanks to Robert Brazile for pointing out that the Sun 3 version | |
85 | was broken. Dynamic library handling is still broken on Sun 3s | |
86 | under 4.1.1U1, but apparently not 4.1.1. If you have such a machine, | |
87 | use -Bstatic.) | |
88 | ||
89 | Version 2.5 fixed the following bugs: | |
90 | - Removed an explicit call to exit(1) | |
91 | - Fixed calls to GC_printf and GC_err_printf, so the correct number of | |
92 | arguments are always supplied. The OS/2 C compiler gets confused if | |
93 | the number of actuals and the number of formals differ. (ANSI C | |
94 | doesn't require this to work. The ANSI sanctioned way of doing things | |
95 | causes too many compatibility problems.) | |
96 | ||
97 | Version 3.0 added generational/incremental collection and stubborn | |
98 | objects. | |
99 | ||
100 | Version 3.1 added the following features: | |
101 | - A workaround for a SunOS 4.X SPARC C compiler | |
102 | misfeature that caused problems when the collector was turned into | |
103 | a dynamic library. | |
104 | - A fix for a bug in GC_base that could result in a memory fault. | |
105 | - A fix for a performance bug (and several other misfeatures) pointed | |
106 | out by Dave Detlefs and Al Dosser. | |
107 | - Use of dirty bit information for static data under Solaris 2.X. | |
108 | - DEC Alpha/OSF1 support (thanks to Al Dosser). | |
109 | - Incremental collection on more platforms. | |
110 | - A more refined heap expansion policy. Less space usage by default. | |
111 | - Various minor enhancements to reduce space usage, and to reduce | |
112 | the amount of memory scanned by the collector. | |
113 | - Uncollectable allocation without per object overhead. | |
114 | - More conscientious handling of out-of-memory conditions. | |
115 | - Fixed a bug in debugging stubborn allocation. | |
116 | - Fixed a bug that resulted in occasional erroneous reporting of smashed | |
117 | objects with debugging allocation. | |
118 | - Fixed bogus leak reports of size 4096 blocks with FIND_LEAK. | |
119 | ||
120 | Version 3.2 fixed a serious and not entirely repeatable bug in | |
121 | the incremental collector. It appeared only when dirty bit info | |
122 | on the roots was available, which is normally only under Solaris. | |
123 | It also added GC_general_register_disappearing_link, and some | |
124 | testing code. Interface.c disappeared. | |
125 | ||
126 | Version 3.3 fixes several bugs and adds new ports: | |
127 | - PCR-specific bugs. | |
128 | - Missing locking in GC_free, redundant FASTUNLOCK | |
129 | in GC_malloc_stubborn, and 2 bugs in | |
130 | GC_unregister_disappearing_link. | |
131 | All of the above were pointed out by Neil Sharman | |
132 | (neil@cs.mu.oz.au). | |
133 | - Common symbols allocated by the SunOS4.X dynamic loader | |
134 | were not included in the root set. | |
135 | - Bug in GC_finalize (reported by Brian Beuning and Al Dosser) | |
136 | - Merged Amiga port from Jesper Peterson (untested) | |
137 | - Merged NeXT port from Thomas Funke (significantly | |
138 | modified and untested) | |
139 | ||
140 | Version 3.4: | |
141 | - Fixed a performance bug in GC_realloc. | |
142 | - Updated the amiga port. | |
143 | - Added NetBSD and 386BSD ports. | |
144 | - Added cord library. | |
145 | - Added trivial performance enhancement for | |
146 | ALL_INTERIOR_POINTERS. (Don't scan last word.) | |
147 | ||
148 | Version 3.5 | |
149 | - Minor collections now mark from roots only once, if that | |
150 | doesn't cause an excessive pause. | |
151 | - The stack clearing heuristic was refined to prevent anomalies | |
152 | with very heavily recursive programs and sparse stacks. | |
153 | - Fixed a bug that prevented mark stack growth in some cases. | |
154 | GC_objects_are_marked should be set to TRUE after a call | |
155 | to GC_push_roots and as part of GC_push_marked, since | |
156 | both can now set mark bits. I think this is only a performance | |
157 | bug, but I wouldn't bet on it. It's certainly very hard to argue | |
158 | that the old version was correct. | |
159 | - Fixed an incremental collection bug that prevented it from | |
160 | working at all when HBLKSIZE != getpagesize() | |
161 | - Changed dynamic_loading.c to include gc_priv.h before testing | |
162 | DYNAMIC_LOADING. SunOS dynamic library scanning | |
163 | must have been broken in 3.4. | |
164 | - Object size rounding now adapts to program behavior. | |
165 | - Added a workaround (provided by Manuel Serrano and | |
166 | colleagues) to a long-standing SunOS 4.X (and 3.X?) ld bug | |
167 | that I had incorrectly assumed to have been squished. | |
168 | The collector was broken if the text segment size was within | |
169 | 32 bytes of a multiple of 8K bytes, and if the beginning of | |
170 | the data segment contained interesting roots. The workaround | |
171 | assumes a demand-loadable executable. The original may have | |
172 | have "worked" in some other cases. | |
173 | - Added dynamic library support under IRIX5. | |
174 | - Added support for EMX under OS/2 (thanks to Ari Huttunen). | |
175 | ||
176 | Version 3.6: | |
177 | - fixed a bug in the mark stack growth code that was introduced | |
178 | in 3.4. | |
179 | - fixed Makefile to work around DEC AXP compiler tail recursion | |
180 | bug. | |
181 | ||
182 | Version 3.7: | |
183 | - Added a workaround for an HP/UX compiler bug. | |
184 | - Fixed another stack clearing performance bug. Reworked | |
185 | that code once more. | |
186 | ||
187 | Version 4.0: | |
188 | - Added support for Solaris threads (which was possible | |
189 | only by reimplementing some fraction of Solaris threads, | |
190 | since Sun doesn't currently make the thread debugging | |
191 | interface available). | |
192 | - Added non-threads win32 and win32S support. | |
193 | - (Grudgingly, with suitable muttering of obscenities) renamed | |
194 | files so that the collector distribution could live on a FAT | |
195 | file system. Files that are guaranteed to be useless on | |
196 | a PC still have long names. Gc_inline.h and gc_private.h | |
197 | still exist, but now just include gc_inl.h and gc_priv.h. | |
198 | - Fixed a really obscure bug in finalization that could cause | |
199 | undetected mark stack overflows. (I would be surprised if | |
200 | any real code ever tickled this one.) | |
201 | - Changed finalization code to dynamically resize the hash | |
202 | tables it maintains. (This probably does not matter for well- | |
203 | -written code. It no doubt does for C++ code that overuses | |
204 | destructors.) | |
205 | - Added typed allocation primitives. Rewrote the marker to | |
206 | accommodate them with more reasonable efficiency. This | |
207 | change should also speed up marking for GC_malloc allocated | |
208 | objects a little. See gc_typed.h for new primitives. | |
209 | - Improved debugging facilities slightly. Allocation time | |
210 | stack traces are now kept by default on SPARC/SUNOS4. | |
211 | (Thanks to Scott Schwartz.) | |
212 | - Added better support for small heap applications. | |
213 | - Significantly extended cord package. Fixed a bug in the | |
214 | implementation of lazily read files. Printf and friends now | |
215 | have cord variants. Cord traversals are a bit faster. | |
216 | - Made ALL_INTERIOR_POINTERS recognition the default. | |
217 | - Fixed de so that it can run in constant space, independent | |
218 | of file size. Added simple string searching to cords and de. | |
219 | - Added the Hull-Ellis C++ interface. | |
220 | - Added dynamic library support for OSF/1. | |
221 | (Thanks to Al Dosser and Tim Bingham at DEC.) | |
222 | - Changed argument to GC_expand_hp to be expressed | |
223 | in units of bytes instead of heap blocks. (Necessary | |
224 | since the heap block size now varies depending on | |
225 | configuration. The old version was never very clean.) | |
226 | - Added GC_get_heap_size(). The previous "equivalent" | |
227 | was broken. | |
228 | - Restructured the Makefile a bit. | |
229 | ||
230 | Since version 4.0: | |
231 | - Changed finalization implementation to guarantee that | |
232 | finalization procedures are called outside of the allocation | |
233 | lock, making direct use of the interface a little less dangerous. | |
234 | MAY BREAK EXISTING CLIENTS that assume finalizers | |
235 | are protected by a lock. Since there seem to be few multithreaded | |
236 | clients that use finalization, this is hopefully not much of | |
237 | a problem. | |
238 | - Fixed a gross bug in CORD_prev. | |
239 | - Fixed a bug in blacklst.c that could result in unbounded | |
240 | heap growth during startup on machines that do not clear | |
241 | memory obtained from the OS (e.g. win32S). | |
242 | - Ported de editor to win32/win32S. (This is now the only | |
243 | version with a mouse-sensitive UI.) | |
244 | - Added GC_malloc_ignore_off_page to allocate large arrays | |
245 | in the presence of ALL_INTERIOR_POINTERS. | |
246 | - Changed GC_call_with_alloc_lock to not disable signals in | |
247 | the single-threaded case. | |
248 | - Reduced retry count in GC_collect_or_expand for garbage | |
249 | collecting when out of memory. | |
250 | - Made uncollectable allocations bypass black-listing, as they | |
251 | should. | |
252 | - Fixed a bug in typed_test in test.c that could cause (legitimate) | |
253 | GC crashes. | |
254 | - Fixed some potential synchronization problems in finalize.c | |
255 | - Fixed a real locking problem in typd_mlc.c. | |
256 | - Worked around an AIX 3.2 compiler feature that results in | |
257 | out of bounds memory references. | |
258 | - Partially worked around an IRIX5.2 beta problem (which may | |
259 | or may not persist to the final release). | |
260 | - Fixed a bug in the heap integrity checking code that could | |
261 | result in explicitly deallocated objects being identified as | |
262 | smashed. Fixed a bug in the dbg_mlc stack saving code | |
263 | that caused old argument pointers to be considered live. | |
264 | - Fixed a bug in CORD_ncmp (and hence CORD_str). | |
265 | - Repaired the OS2 port, which had suffered from bit rot | |
266 | in 4.0. Worked around what appears to be CSet/2 V1.0 | |
267 | optimizer bug. | |
268 | - Fixed a Makefile bug for target "c++". | |
269 | ||
270 | Since version 4.1: | |
271 | - Multiple bug fixes/workarounds in the Solaris threads version. | |
272 | (It occasionally failed to locate some register contents for | |
273 | marking. It also turns out that thr_suspend and friends are | |
274 | unreliable in Solaris 2.3. Dirty bit reads appear | |
275 | to be unreliable under some weird | |
276 | circumstances. My stack marking code | |
277 | contained a serious performance bug. The new code is | |
278 | extremely defensive, and has not failed in several cpu | |
279 | hours of testing. But no guarantees ...) | |
280 | - Added MacOS support (thanks to Patrick Beard.) | |
281 | - Fixed several syntactic bugs in gc_c++.h and friends. (These | |
282 | didn't bother g++, but did bother most other compilers.) | |
283 | Fixed gc_c++.h finalization interface. (It didn't.) | |
284 | - 64 bit alignment for allocated objects was not guaranteed in a | |
285 | few cases in which it should have been. | |
286 | - Added GC_malloc_atomic_ignore_off_page. | |
287 | - Added GC_collect_a_little. | |
288 | - Added some prototypes to gc.h. | |
289 | - Some other minor bug fixes (notably in Makefile). | |
290 | - Fixed OS/2 / EMX port (thanks to Ari Huttunen). | |
291 | - Fixed AmigaDOS port. (thanks to Michel Schinz). | |
292 | - Fixed the DATASTART definition under Solaris. There | |
293 | was a 1 in 16K chance of the collector missing the first | |
294 | 64K of static data (and thus crashing). | |
295 | - Fixed some blatant anachronisms in the README file. | |
296 | - Fixed PCR-Makefile for upcoming PPCR release. | |
297 | ||
298 | Since version 4.2: | |
299 | - Fixed SPARC alignment problem with GC_DEBUG. | |
300 | - Fixed Solaris threads /proc workaround. The real | |
301 | problem was an interaction with mprotect. | |
302 | - Incorporated fix from Patrick Beard for gc_c++.h (now gc_cpp.h). | |
303 | - Slightly improved allocator space utilization by | |
304 | fixing the GC_size_map mechanism. | |
305 | - Integrated some Sony News and MIPS RISCos 4.51 | |
306 | patches. (Thanks to Nobuyuki Hikichi of | |
307 | Software Research Associates, Inc. Japan) | |
308 | - Fixed HP_PA alignment problem. (Thanks to | |
309 | xjam@cork.cs.berkeley.edu.) | |
310 | - Added GC_same_obj and friends. Changed GC_base | |
311 | to return 0 for pointers past the end of large objects. | |
312 | Improved GC_base performance with ALL_INTERIOR_POINTERS | |
313 | on machines with a slow integer mod operation. | |
314 | Added GC_PTR_ADD, GC_PTR_STORE, etc. to prepare | |
315 | for preprocessor. | |
316 | - changed the default on most UNIX machines to be that | |
317 | signals are not disabled during critical GC operations. | |
318 | This is still ANSI-conforming, though somewhat dangerous | |
319 | in the presence of signal handlers. But the performance | |
320 | cost of the alternative is sometimes problematic. | |
321 | Can be changed back with a minor Makefile edit. | |
322 | - renamed IS_STRING in gc.h, to CORD_IS_STRING, thus | |
323 | following my own naming convention. Added the function | |
324 | CORD_to_const_char_star. | |
325 | - Fixed a gross bug in GC_finalize. Symptom: occasional | |
326 | address faults in that function. (Thanks to Anselm | |
327 | Baird-Smith (Anselm.BairdSmith@inria.fr) | |
328 | - Added port to ICL DRS6000 running DRS/NX. Restructured | |
329 | things a bit to factor out common code, and remove obsolete | |
330 | code. Collector should now run under SUNOS5 with either | |
331 | mprotect or /proc dirty bits. (Thanks to Douglas Steel | |
332 | (doug@wg.icl.co.uk)). | |
333 | - More bug fixes and workarounds for Solaris 2.X. (These were | |
334 | mostly related to putting the collector in a dynamic library, | |
335 | which didn't really work before. Also SOLARIS_THREADS | |
336 | didn't interact well with dl_open.) Thanks to btlewis@eng.sun.com. | |
337 | - Fixed a serious performance bug on the DEC Alpha. The text | |
338 | segment was getting registered as part of the root set. | |
339 | (Amazingly, the result was still fast enough that the bug | |
340 | was not conspicuous.) The fix works on OSF/1, version 1.3. | |
341 | Hopefully it also works on other versions of OSF/1 ... | |
342 | - Fixed a bug in GC_clear_roots. | |
343 | - Fixed a bug in GC_generic_malloc_words_small that broke | |
344 | gc_inl.h. (Reported by Antoine de Maricourt. I broke it | |
345 | in trying to tweak the Mac port.) | |
346 | - Fixed some problems with cord/de under Linux. | |
347 | - Fixed some cord problems, notably with CORD_riter4. | |
348 | - Added DG/UX port. | |
349 | Thanks to Ben A. Mesander (ben@piglet.cr.usgs.gov) | |
350 | - Added finalization registration routines with weaker ordering | |
351 | constraints. (This is necessary for C++ finalization with | |
352 | multiple inheritance, since the compiler often adds self-cycles.) | |
353 | - Filled the holes in the SCO port. (Thanks to Michael Arnoldus | |
354 | <chime@proinf.dk>.) | |
355 | - John Ellis' additions to the C++ support: From John: | |
356 | ||
357 | * I completely rewrote the documentation in the interface gc_c++.h | |
358 | (later renamed gc_cpp.h). I've tried to make it both clearer and more | |
359 | precise. | |
360 | ||
361 | * The definition of accessibility now ignores pointers from an | |
362 | finalizable object (an object with a clean-up function) to itself. | |
363 | This allows objects with virtual base classes to be finalizable by the | |
364 | collector. Compilers typically implement virtual base classes using | |
365 | pointers from an object to itself, which under the old definition of | |
366 | accessibility prevented objects with virtual base classes from ever | |
367 | being collected or finalized. | |
368 | ||
369 | * gc_cleanup now includes gc as a virtual base. This was enabled by | |
370 | the change in the definition of accessibility. | |
371 | ||
372 | * I added support for operator new[]. Since most (all?) compilers | |
373 | don't yet support operator new[], it is conditionalized on | |
374 | -DOPERATOR_NEW_ARRAY. The code is untested, but its trivial and looks | |
375 | correct. | |
376 | ||
377 | * The test program test_gc_c++ (later renamed test_cpp.cc) | |
378 | tries to test for the C++-specific functionality not tested by the | |
379 | other programs. | |
380 | - Added <unistd.h> include to misc.c. (Needed for ppcr.) | |
381 | - Added PowerMac port. (Thanks to Patrick Beard again.) | |
382 | - Fixed "srcdir"-related Makefile problems. Changed things so | |
383 | that all externally visible include files always appear in the | |
384 | include subdirectory of the source. Made gc.h directly | |
385 | includable from C++ code. (These were at Per | |
386 | Bothner's suggestion.) | |
387 | - Changed Intel code to also mark from ebp (Kevin Warne's | |
388 | suggestion). | |
389 | - Renamed C++ related files so they could live in a FAT | |
390 | file system. (Charles Fiterman's suggestion.) | |
391 | - Changed Windows NT Makefile to include C++ support in | |
392 | gc.lib. Added C++ test as Makefile target. | |
393 | ||
394 | Since version 4.3: | |
395 | - ASM_CLEAR_CODE was erroneously defined for HP | |
396 | PA machines, resulting in a compile error. | |
397 | - Fixed OS/2 Makefile to create a library. (Thanks to | |
398 | Mark Boulter (mboulter@vnet.ibm.com)). | |
399 | - Gc_cleanup objects didn't work if they were created on | |
400 | the stack. Fixed. | |
401 | - One copy of Gc_cpp.h in the distribution was out of | |
402 | synch, and failed to document some known compiler | |
403 | problems with explicit destructor invocation. Partially | |
404 | fixed. There are probably other compilers on which | |
405 | gc_cleanup is miscompiled. | |
406 | - Fixed Makefile to pass C compiler flags to C++ compiler. | |
407 | - Added Mac fixes. | |
408 | - Fixed os_dep.c to work around what appears to be | |
409 | a new and different VirtualQuery bug under newer | |
410 | versions of win32S. | |
411 | - GC_non_gc_bytes was not correctly maintained by | |
412 | GC_free. Fixed. Thanks to James Clark (jjc@jclark.com). | |
413 | - Added GC_set_max_heap_size. | |
414 | - Changed allocation code to ignore blacklisting if it is preventing | |
415 | use of a very large block of memory. This has the advantage | |
416 | that naive code allocating very large objects is much more | |
417 | likely to work. The downside is you might no | |
418 | longer find out that such code should really use | |
419 | GC_malloc_ignore_off_page. | |
420 | - Changed GC_printf under win32 to close and reopen the file | |
421 | between calls. FAT file systems otherwise make the log file | |
422 | useless for debugging. | |
423 | - Added GC_try_to_collect and GC_get_bytes_since_gc. These | |
424 | allow starting an abortable collection during idle times. | |
425 | This facility does not require special OS support. (Thanks to | |
426 | Michael Spertus of Geodesic Systems for suggesting this. It was | |
427 | actually an easy addition. Kumar Srikantan previously added a similar | |
428 | facility to a now ancient version of the collector. At the time | |
429 | this was much harder, and the result was less convincing.) | |
430 | - Added some support for the Borland development environment. (Thanks | |
431 | to John Ellis and Michael Spertus.) | |
432 | - Removed a misfeature from checksums.c that caused unexpected | |
433 | heap growth. (Thanks to Scott Schwartz.) | |
434 | - Changed finalize.c to call WARN if it encounters a finalization cycle. | |
435 | WARN is defined in gc_priv.h to write a message, usually to stdout. | |
436 | In many environments, this may be inappropriate. | |
437 | - Renamed NO_PARAMS in gc.h to GC_NO_PARAMS, thus adhering to my own | |
438 | naming convention. | |
439 | - Added GC_set_warn_proc to intercept warnings. | |
440 | - Fixed Amiga port. (Thanks to Michel Schinz (schinz@alphanet.ch).) | |
441 | - Fixed a bug in mark.c that could result in an access to unmapped | |
442 | memory from GC_mark_from_mark_stack on machines with unaligned | |
443 | pointers. | |
444 | - Fixed a win32 specific performance bug that could result in scanning of | |
445 | objects allocated with the system malloc. | |
446 | - Added REDIRECT_MALLOC. | |
447 | ||
448 | Since version 4.4: | |
449 | - Fixed many minor and one major README bugs. (Thanks to Franklin Chen | |
450 | (chen@adi.com) for pointing out many of them.) | |
451 | - Fixed ALPHA/OSF/1 dynamic library support. (Thanks to Jonathan Bachrach | |
452 | (jonathan@harlequin.com)). | |
453 | - Added incremental GC support (MPROTECT_VDB) for Linux (with some | |
454 | help from Bruno Haible). | |
455 | - Altered SPARC recognition tests in gc.h and config.h (mostly as | |
456 | suggested by Fergus Henderson). | |
457 | - Added basic incremental GC support for win32, as implemented by | |
458 | Windows NT and Windows 95. GC_enable_incremental is a noop | |
459 | under win32s, which doesn't implement enough of the VM interface. | |
460 | - Added -DLARGE_CONFIG. | |
461 | - Fixed GC_..._ignore_off_page to also function without | |
462 | -DALL_INTERIOR_POINTERS. | |
463 | - (Hopefully) fixed RS/6000 port. (Only the test was broken.) | |
464 | - Fixed a performance bug in the nonincremental collector running | |
465 | on machines supporting incremental collection with MPROTECT_VDB | |
466 | (e.g. SunOS 4, DEC AXP). This turned into a correctness bug under | |
467 | win32s with win32 incremental collection. (Not all memory protection | |
468 | was disabled.) | |
469 | - Fixed some ppcr related bit rot. | |
470 | - Caused dynamic libraries to be unregistered before reregistering. | |
471 | The old way turned out to be a performance bug on some machines. | |
472 | - GC_root_size was not properly maintained under MSWIN32. | |
473 | - Added -DNO_DEBUGGING and GC_dump. | |
474 | - Fixed a couple of bugs arising with SOLARIS_THREADS + | |
475 | REDIRECT_MALLOC. | |
476 | - Added NetBSD/M68K port. (Thanks to Peter Seebach | |
477 | <seebs@taniemarie.solon.com>.) | |
478 | - Fixed a serious realloc bug. For certain object sizes, the collector | |
479 | wouldn't scan the expanded part of the object. (Thanks to Clay Spence | |
480 | (cds@peanut.sarnoff.com) for noticing the problem, and helping me to | |
481 | track it down.) | |
482 | ||
483 | Since version 4.5: | |
484 | - Added Linux ELF support. (Thanks to Arrigo Triulzi <arrigo@ic.ac.uk>.) | |
485 | - GC_base crashed if it was called before any other GC_ routines. | |
486 | This could happen if a gc_cleanup object was allocated outside the heap | |
487 | before any heap allocation. | |
488 | - The heap expansion heuristic was not stable if all objects had finalization | |
489 | enabled. Fixed finalize.c to count memory in finalization queue and | |
490 | avoid explicit deallocation. Changed alloc.c to also consider this count. | |
491 | (This is still not recommended. It's expensive if nothing else.) Thanks | |
492 | to John Ellis for pointing this out. | |
493 | - GC_malloc_uncollectable(0) was broken. Thanks to Phong Vo for pointing | |
494 | this out. | |
495 | - The collector didn't compile under Linux 1.3.X. (Thanks to Fred Gilham for | |
496 | pointing this out.) The current workaround is ugly, but expected to be | |
497 | temporary. | |
498 | - Fixed a formatting problem for SPARC stack traces. | |
499 | - Fixed some '=='s in os_dep.c that should have been assignments. | |
500 | Fortunately these were in code that should never be executed anyway. | |
501 | (Thanks to Fergus Henderson.) | |
502 | - Fixed the heap block allocator to only drop blacklisted blocks in small | |
503 | chunks. Made BL_LIMIT self adjusting. (Both of these were in response | |
504 | to heap growth observed by Paul Graham.) | |
505 | - Fixed the Metrowerks/68K Mac code to also mark from a6. (Thanks | |
506 | to Patrick Beard.) | |
507 | - Significantly updated README.debugging. | |
508 | - Fixed some problems with longjmps out of signal handlers, especially under | |
509 | Solaris. Added a workaround for the fact that siglongjmp doesn't appear to | |
510 | do the right thing with -lthread under Solaris. | |
511 | - Added MSDOS/djgpp port. (Thanks to Mitch Harris (maharri@uiuc.edu).) | |
512 | - Added "make reserved_namespace" and "make user_namespace". The | |
513 | first renames ALL "GC_xxx" identifiers as "_GC_xxx". The second is the | |
514 | inverse transformation. Note that doing this is guaranteed to break all | |
515 | clients written for the other names. | |
516 | - descriptor field for kind NORMAL in GC_obj_kinds with ADD_BYTE_AT_END | |
517 | defined should be -ALIGNMENT not WORDS_TO_BYTES(-1). This is | |
518 | a serious bug on machines with pointer alignment of less than a word. | |
519 | - GC_ignore_self_finalize_mark_proc didn't handle pointers to very near the | |
520 | end of the object correctly. Caused failures of the C++ test on a DEC Alpha | |
521 | with g++. | |
522 | - gc_inl.h still had problems. Partially fixed. Added warnings at the | |
523 | beginning to hopefully specify the remaining dangers. | |
524 | - Added DATAEND definition to config.h. | |
525 | - Fixed some of the .h file organization. Fixed "make floppy". | |
526 | ||
527 | Since version 4.6: | |
528 | - Fixed some compilation problems with -DCHECKSUMS (thanks to Ian Searle) | |
529 | - Updated some Mac specific files to synchronize with Patrick Beard. | |
530 | - Fixed a serious bug for machines with non-word-aligned pointers. | |
531 | (Thanks to Patrick Beard for pointing out the problem. The collector | |
532 | should fail almost any conceivable test immediately on such machines.) | |
533 | ||
534 | Since version 4.7: | |
535 | - Changed a "comment" in a MacOS specific part of mach-dep.c that caused | |
536 | gcc to fail on other platforms. | |
537 | ||
538 | Since version 4.8 | |
539 | - More README.debugging fixes. | |
540 | - Objects ready for finalization, but not finalized in the same GC | |
541 | cycle, could be prematurely collected. This occasionally happened | |
542 | in test_cpp. | |
543 | - Too little memory was obtained from the system for very large | |
544 | objects. That could cause a heap explosion if these objects were | |
545 | not contiguous (e.g. under PCR), and too much of them was blacklisted. | |
546 | - Due to an improper initialization, the collector was too hesitant to | |
547 | allocate blacklisted objects immediately after system startup. | |
548 | - Moved GC_arrays from the data into the bss segment by not explicitly | |
549 | initializing it to zero. This significantly | |
550 | reduces the size of executables, and probably avoids some disk accesses | |
551 | on program startup. It's conceivable that it might break a port that I | |
552 | didn't test. | |
553 | - Fixed EMX_MAKEFILE to reflect the gc_c++.h to gc_cpp.h renaming which | |
554 | occurred a while ago. | |
555 | ||
556 | Since 4.9: | |
557 | - Fixed a typo around a call to GC_collect_or_expand in alloc.c. It broke | |
558 | handling of out of memory. (Thanks to Patrick Beard for noticing.) | |
559 | ||
560 | Since 4.10: | |
561 | - Rationalized (hopefully) GC_try_to_collect in an incremental collection | |
562 | environment. It appeared to not handle a call while a collection was in | |
563 | progress, and was otherwise too conservative. | |
564 | - Merged GC_reclaim_or_delete_all into GC_reclaim_all to get rid of some | |
565 | code. | |
566 | - Added Patrick Beard's Mac fixes, with substantial completely untested | |
567 | modifications. | |
568 | - Fixed the MPROTECT_VDB code to deal with large pages and imprecise | |
569 | fault addresses (as on an UltraSPARC running Solaris 2.5). Note that this | |
570 | was not a problem in the default configuration, which uses PROC_VDB. | |
571 | - The DEC Alpha assembly code needed to restore $gp between calls. | |
572 | Thanks to Fergus Henderson for tracking this down and supplying a | |
573 | patch. | |
574 | - The write command for "de" was completely broken for large files. | |
575 | I used the easiest portable fix, which involved changing the semantics | |
576 | so that f.new is written instead of overwriting f. That's safer anyway. | |
577 | - Added README.solaris2 with a discussion of the possible problems of | |
578 | mixing the collector's sbrk allocation with malloc/realloc. | |
579 | - Changed the data segment starting address for SGI machines. The | |
580 | old code failed under IRIX6. | |
581 | - Required double word alignment for MIPS. | |
582 | - Various minor fixes to remove warnings. | |
583 | - Attempted to fix some Solaris threads problems reported by Zhiying Chen. | |
584 | In particular, the collector could try to fork a thread with the | |
585 | world stopped as part of GC_thr_init. It also failed to deal with | |
586 | the case in which the original thread terminated before the whole | |
587 | process did. | |
588 | - Added -DNO_EXECUTE_PERMISSION. This has a major performance impact | |
589 | on the incremental collector under Irix, and perhaps under other | |
590 | operating systems. | |
591 | - Added some code to support allocating the heap with mmap. This may | |
592 | be preferable under some circumstances. | |
593 | - Integrated dynamic library support for HP. | |
594 | (Thanks to Knut Tvedten <knuttv@ifi.uio.no>.) | |
595 | - Integrated James Clark's win32 threads support, and made a number | |
596 | of changes to it, many of which were suggested by Pontus Rydin. | |
597 | This is still not 100% solid. | |
598 | - Integrated Alistair Crooks' support for UTS4 running on an Amdahl | |
599 | 370-class machine. | |
600 | - Fixed a serious bug in explicitly typed allocation. Objects requiring | |
601 | large descriptors where handled in a way that usually resulted in | |
602 | a segmentation fault in the marker. (Thanks to Jeremy Fitzhardinge | |
603 | for helping to track this down.) | |
604 | - Added partial support for GNU win32 development. (Thanks to Fergus | |
605 | Henderson.) | |
606 | - Added optional support for Java-style finalization semantics. (Thanks | |
607 | to Patrick Bridges.) This is recommended only for Java implementations. | |
608 | - GC_malloc_uncollectable faulted instead of returning 0 when out of | |
609 | memory. (Thanks to dan@math.uiuc.edu for noticing.) | |
610 | - Calls to GC_base before the collector was initialized failed on a | |
611 | DEC Alpha. (Thanks to Matthew Flatt.) | |
612 | - Added base pointer checking to GC_REGISTER_FINALIZER in debugging | |
613 | mode, at the suggestion of Jeremy Fitzhardinge. | |
614 | - GC_debug_realloc failed for uncollectable objects. (Thanks to | |
615 | Jeremy Fitzhardinge.) | |
616 | - Explicitly typed allocation could crash if it ran out of memory. | |
617 | (Thanks to Jeremy Fitzhardinge.) | |
618 | - Added minimal support for a DEC Alpha running Linux. | |
619 | - Fixed a problem with allocation of objects whose size overflowed | |
620 | ptrdiff_t. (This now fails unconditionally, as it should.) | |
621 | - Added the beginning of Irix pthread support. | |
622 | - Integrated Xiaokun Zhu's fixes for djgpp 2.01. | |
623 | - Added SGI-style STL allocator support (gc_alloc.h). | |
624 | - Fixed a serious bug in README.solaris2. Multithreaded programs must include | |
625 | gc.h with SOLARIS_THREADS defined. | |
626 | - Changed GC_free so it actually deallocates uncollectable objects. | |
627 | (Thanks to Peter Chubb for pointing out the problem.) | |
628 | - Added Linux ELF support for dynamic libararies. (Thanks again to | |
629 | Patrick Bridges.) | |
630 | - Changed the Borland cc configuration so that the assembler is not | |
631 | required. | |
632 | - Fixed a bug in the C++ test that caused it to fail in 64-bit | |
633 | environments. | |
634 | ||
635 | Since 4.11: | |
636 | - Fixed ElfW definition in dyn_load.c. (Thanks to Fergus Henderson.) | |
637 | This prevented the dynamic library support from compiling on some | |
638 | older ELF Linux systems. | |
639 | - Fixed UTS4 port (which I apparently mangled during the integration) | |
640 | (Thanks to again to Alistair Crooks.) | |
641 | - "Make C++" failed on Suns with SC4.0, due to a problem with "bool". | |
642 | Fixed in gc_priv.h. | |
643 | - Added more pieces for GNU win32. (Thanks to Timothy N. Newsham.) | |
644 | The current state of things should suffice for at least some | |
645 | applications. | |
646 | - Changed the out of memory retry count handling as suggested by | |
647 | Kenjiro Taura. (This matters only if GC_max_retries > 0, which | |
648 | is no longer the default.) | |
649 | - If a /proc read failed repeatedly, GC_written_pages was not updated | |
650 | correctly. (Thanks to Peter Chubb for diagnosing this.) | |
651 | - Under unlikely circumstances, the allocator could infinite loop in | |
652 | an out of memory situation. (Thanks again to Kenjiro Taura for | |
653 | identifying the problem and supplying a fix.) | |
654 | - Fixed a syntactic error in the DJGPP code. (Thanks to Fergus | |
655 | Henderson for finding this by inspection.) Also fixed a test program | |
656 | problem with DJGPP (Thanks to Peter Monks.) | |
657 | - Atomic uncollectable objects were not treated correctly by the | |
658 | incremental collector. This resulted in weird log statistics and | |
659 | occasional performance problems. (Thanks to Peter Chubb for pointing | |
660 | this out.) | |
661 | - Fixed some problems resulting from compilers that dont define | |
662 | __STDC__. In this case void * and char * were used inconsistently | |
663 | in some cases. (Void * should not have been used at all. If | |
664 | you have an ANSI superset compiler that does not define __STDC__, | |
665 | please compile with -D__STDC__=0. Thanks to Manuel Serrano and others | |
666 | for pointing out the problem.) | |
667 | - Fixed a compilation problem on Irix with -n32 and -DIRIX_THREADS. | |
668 | Also fixed some other IRIX_THREADS problems which may or may not have | |
669 | had observable symptoms. | |
670 | - Fixed an HP PA compilation problem in dyn_load.c. (Thanks to | |
671 | Philippe Queinnec.) | |
672 | - SEGV fault handlers sometimes did not get reset correctly. (Thanks | |
673 | to David Pickens.) | |
674 | - Added a fix for SOLARIS_THREADS on Intel. (Thanks again to David | |
675 | Pickens.) This probably needs more work to become functional. | |
676 | - Fixed struct sigcontext_struct in os_dep.c for compilation under | |
677 | Linux 2.1.X. (Thanks to Fergus Henderson.) | |
678 | - Changed the DJGPP STACKBOTTOM and DATASTART values to those suggested | |
679 | by Kristian Kristensen. These may still not be right, but it is | |
680 | it is likely to work more often than what was there before. They may | |
681 | even be exactly right. | |
682 | - Added a #include <string.h> to test_cpp.cc. This appears to help | |
683 | with HP/UX and gcc. (Thanks to assar@sics.se.) | |
684 | - Version 4.11 failed to run in incremental mode on recent 64-bit Irix | |
685 | kernels. This was a problem related to page unaligned heap segments. | |
686 | Changed the code to page align heap sections on all platforms. | |
687 | (I had mistakenly identified this as a kernel problem earlier. | |
688 | It was not.) | |
689 | - Version 4.11 did not make allocated storage executable, except on | |
690 | one or two platforms, due to a bug in a #if test. (Thanks to Dave | |
691 | Grove for pointing this out.) | |
692 | - Added sparc_sunos4_mach_dep.s to support Sun's compilers under SunOS4. | |
693 | - Added GC_exclude_static_roots. | |
694 | - Fixed the object size mapping algorithm. This shouldn't matter, | |
695 | but the old code was ugly. | |
696 | - Heap checking code could die if one of the allocated objects was | |
697 | larger than its base address. (Unsigned underflow problem. Thanks | |
698 | to Clay Spence for isolating the problem.) | |
699 | - Added RS6000 (AIX) dynamic library support and fixed STACK_BOTTOM. | |
700 | (Thanks to Fred Stearns.) | |
701 | - Added Fergus Henderson's patches for improved robustness with large | |
702 | heaps and lots of blacklisting. | |
703 | - Added Peter Chubb's changes to support Solaris Pthreads, to support | |
704 | MMAP allocation in Solaris, to allow Solaris to find dynamic libraries | |
705 | through /proc, to add malloc_typed_ignore_off_page, and a few other | |
706 | minor features and bug fixes. | |
707 | - The Solaris 2 port should not use sbrk. I received confirmation from | |
708 | Sun that the use of sbrk and malloc in the same program is not | |
709 | supported. The collector now defines USE_MMAP by default on Solaris. | |
710 | - Replaced the djgpp makefile with Gary Leavens' version. | |
711 | - Fixed MSWIN32 detection test. | |
712 | - Added Fergus Henderson's patches to allow putting the collector into | |
713 | a DLL under GNU win32. | |
714 | - Added Ivan V. Demakov's port to Watcom C on X86. | |
715 | - Added Ian Piumarta's Linux/PowerPC port. | |
716 | - On Brian Burton's suggestion added PointerFreeGC to the placement | |
717 | options in gc_cpp.h. This is of course unsafe, and may be controversial. | |
718 | On the other hand, it seems to be needed often enough that it's worth | |
719 | adding as a standard facility. | |
720 | ||
721 | Since 4.12: | |
722 | - Fixed a crucial bug in the Watcom port. There was a redundant decl | |
723 | of GC_push_one in gc_priv.h. | |
724 | - Added FINALIZE_ON_DEMAND. | |
725 | - Fixed some pre-ANSI cc problems in test.c. | |
726 | - Removed getpagesize() use for Solaris. It seems to be missing in one | |
727 | or two versions. | |
728 | - Fixed bool handling for SPARCCompiler version 4.2. | |
729 | - Fixed some files in include that had gotten unlinked from the main | |
730 | copy. | |
731 | - Some RS/6000 fixes (missing casts). Thanks to Toralf Foerster. | |
732 | - Fixed several problems in GC_debug_realloc, affecting mostly the | |
733 | FIND_LEAK case. | |
734 | - GC_exclude_static_roots contained a buggy unsigned comparison to | |
735 | terminate a loop. (Thanks to Wilson Ho.) | |
736 | - CORD_str failed if the substring occurred at the last possible position. | |
737 | (Only affects cord users.) | |
738 | - Fixed Linux code to deal with RedHat 5.0 and integrated Peter Bigot's | |
739 | os_dep.c code for dealing with various Linux versions. | |
740 | - Added workaround for Irix pthreads sigaction bug and possible signal | |
741 | misdirection problems. | |
742 | Since alpha1: | |
743 | - Changed RS6000 STACKBOTTOM. | |
744 | - Integrated Patrick Beard's Mac changes. | |
745 | - Alpha1 didn't compile on Irix m.n, m < 6. | |
746 | - Replaced Makefile.dj with a new one from Gary Leavens. | |
747 | - Added Andrew Stitcher's changes to support SCO OpenServer. | |
748 | - Added PRINT_BLACK_LIST, to allow debugging of high densities of false | |
749 | pointers. | |
750 | - Added code to debug allocator to keep track of return address | |
751 | in GC_malloc caller, thus giving a bit more context. | |
752 | - Changed default behavior of large block allocator to more | |
753 | aggressively avoid fragmentation. This is likely to slow down the | |
754 | collector when it succeeds at reducing space cost. | |
755 | - Integrated Fergus Henderson's CYGWIN32 changes. They are untested, | |
756 | but needed for newer versions. | |
757 | - USE_MMAP had some serious bugs. This caused the collector to fail | |
758 | consistently on Solaris with -DSMALL_CONFIG. | |
759 | - Added Linux threads support, thanks largely to Fergus Henderson. | |
760 | Since alpha2: | |
761 | - Fixed more Linux threads problems. | |
762 | - Changed default GC_free_space_divisor to 3 with new large block allocation. | |
763 | (Thanks to Matthew Flatt for some measurements that suggest the old | |
764 | value sometimes favors space too much over time.) | |
765 | - More CYGWIN32 fixes. | |
766 | - Integrated Tyson-Dowd's Linux-M68K port. | |
767 | - Minor HP PA and DEC UNIX fixes from Fergus Henderson. | |
768 | - Integrated Christoffe Raffali's Linux-SPARC changes. | |
769 | - Allowed for one more GC fixup iteration after a full GC in incremental | |
770 | mode. Some quick measurements suggested that this significantly | |
771 | reduces pause times even with smaller GC_RATE values. | |
772 | - Moved some more GC data structures into GC_arrays. This decreases | |
773 | pause times and GC overhead, but makes debugging slightly less convenient. | |
774 | - Fixed namespace pollution problem ("excl_table"). | |
775 | - Made GC_incremental a constant for -DSMALL_CONFIG, hopefully shrinking | |
776 | that slightly. | |
777 | - Added some win32 threads fixes. | |
778 | - Integrated Ivan Demakov and David Stes' Watcom fixes. | |
779 | - Various other minor fixes contributed by many people. | |
780 | - Renamed config.h to gcconfig.h, since config.h tends to be used for | |
781 | many other things. | |
782 | - Integrated Matthew Flatt's support for 68K MacOS "far globals". | |
783 | - Fixed up some of the dynamic library Makefile targets for consistency | |
784 | across platforms. | |
785 | - Fixed a USE_MMAP typo that caused out-of-memory handling to fail | |
786 | on Solaris. | |
787 | - Added code to test.c to test thread creation a bit more. | |
788 | - Integrated GC_win32_free_heap, as suggested by Ivan Demakov. | |
789 | - Fixed Solaris 2.7 stack base finding problem. (This may actually | |
790 | have been done in an earlier alpha release.) | |
791 | Since alpha3: | |
792 | - Fixed MSWIN32 recognition test, which interfered with cygwin. | |
793 | - Removed unnecessary gc_watcom.asm from distribution. Removed | |
794 | some obsolete README.win32 text. | |
795 | - Added Alpha Linux incremental GC support. (Thanks to Philipp Tomsich | |
796 | for code for retrieving the fault address in a signal handler.) | |
797 | Changed Linux signal handler context argument to be a pointer. | |
798 | - Took care of some new warnings generated by the 7.3 SGI compiler. | |
799 | - Integrated Phillip Musumeci's FreeBSD/ELF fixes. | |
800 | - -DIRIX_THREADS was broken with the -o32 ABI (typo in gc_priv.h> | |
801 | ||
802 | Since 4.13: | |
803 | - Fixed GC_print_source_ptr to not use a prototype. | |
804 | - generalized CYGWIN test. | |
805 | - gc::new did the wrong thing with PointerFreeGC placement. | |
806 | (Thanks to Rauli Ruohonen.) | |
807 | - In the ALL_INTERIOR_POINTERS (default) case, some callee-save register | |
808 | values could fail to be scanned if the register was saved and | |
809 | reused in a GC frame. This showed up in verbose mode with gctest | |
810 | compiled with an unreleased SGI compiler. I vaguely recall an old | |
811 | bug report that may have been related. The bug was probably quite old. | |
812 | (The problem was that the stack scanning could be deferred until | |
813 | after the relevant frame was overwritten, and the new save location | |
814 | might be outside the scanned area. Fixed by more eager stack scanning.) | |
815 | - PRINT_BLACK_LIST had some problems. A few source addresses were garbage. | |
816 | - Replaced Makefile.dj and added -I flags to cord make targets. | |
817 | (Thanks to Gary Leavens.) | |
818 | - GC_try_to_collect was broken with the nonincremental collector. | |
819 | - gc_cleanup destructors could pass the wrong address to | |
820 | GC_register_finalizer_ignore_self in the presence of multiple | |
821 | inheritance. (Thanks to Darrell Schiebel.) | |
822 | - Changed PowerPC Linux stack finding code. | |
823 | ||
824 | Since 4.14alpha1 | |
825 | - -DSMALL_CONFIG did not work reliably with large (> 4K) pages. | |
826 | Recycling the mark stack during expansion could result in a size | |
827 | zero heap segment, which confused things. (This was probably also an | |
828 | issue with the normal config and huge pages.) | |
829 | - Did more work to make sure that callee-save registers were scanned | |
830 | completely, even with the setjmp-based code. Added USE_GENERIC_PUSH_REGS | |
831 | macro to facilitate testing on machines I have access to. | |
832 | - Added code to explicitly push register contents for win32 threads. | |
833 | This seems to be necessary. (Thanks to Pierre de Rop.) | |
834 | ||
835 | Since 4.14alpha2 | |
836 | - changed STACKBOTTOM for DJGPP (Thanks to Salvador Eduardo Tropea). | |
837 | ||
838 | Since 4.14 | |
839 | - Reworked large block allocator. Now uses multiple doubly linked free | |
840 | lists to approximate best fit. | |
841 | - Changed heap expansion heuristic. Entirely free blocks are no longer | |
842 | counted towards the heap size. This seems to have a major impact on | |
843 | heap size stability; the old version could expand the heap way too | |
844 | much in the presence of large block fragmentation. | |
845 | - added -DGC_ASSERTIONS and some simple assertions inside the collector. | |
846 | This is mainlyt for collector debugging. | |
847 | - added -DUSE_MUNMAP to allow the heap to shrink. Suupported on only | |
848 | a few UNIX-like platforms for now. | |
849 | - added GC_dump_regions() for debugging of fragmentation issues. | |
850 | - Changed PowerPC pointer alignment under Linux to 4. (This needs | |
851 | checking by someone who has one. The suggestions came to me via a | |
852 | rather circuitous path.) | |
853 | - Changed the Linux/Alpha port to walk the data segment backwards until | |
854 | it encounters a SIGSEGV. The old way to find the start of the data | |
855 | segment broke with a recent release. | |
856 | - cordxtra.c needed to call GC_REGISTER_FINALIZER instead of | |
857 | GC_register_finalizer, so that it would continue to work with GC_DEBUG. | |
858 | - allochblk sometimes cleared the wrong block for debugging purposes | |
859 | when it dropped blacklisted blocks. This could result in spurious | |
860 | error reports with GC_DEBUG. | |
861 | - added MACOS X Server support. (Thanks to Andrew Stone.) | |
862 | - Changed the Solaris threads code to ignore stack limits > 8 MB with | |
863 | a warning. Empirically, it is not safe to access arbitrary pages | |
864 | in such large stacks. And the dirty bit implementation does not | |
865 | guarantee that none of them will be accessed. | |
866 | - Integrated Martin Tauchmann's Amiga changes. | |
867 | - Integrated James Dominy's OpenBSD/SPARC port. | |
868 | ||
869 | Since 5.0alpha1 | |
870 | - Fixed bugs introduced in alpha1 (OpenBSD & large block initialization). | |
871 | - Added -DKEEP_BACK_PTRS and backptr.h interface. (The implementation | |
872 | idea came from Al Demers.) | |
873 | ||
874 | Since 5.0alpha2 | |
875 | - Added some highly incomplete code to support a copied young generation. | |
876 | Comments on nursery.h are appreciated. | |
877 | - Changed -DFIND_LEAK, -DJAVA_FINALIZATION, and -DFINALIZE_ON_DEMAND, | |
878 | so the same effect could be obtained with a runtime switch. This is | |
879 | a step towards standardizing on a single dynamic GC library. | |
880 | - Significantly changed the way leak detection is handled, as a consequence | |
881 | of the above. | |
882 | ||
883 | Since 5.0 alpha3 | |
884 | - Added protection fault handling patch for Linux/M68K from Fergus | |
885 | Henderson and Roman Hodek. | |
886 | - Removed the tests for SGI_SOURCE in new_gc_alloc.h. This was causing that | |
887 | interface to fail on nonSGI platforms. | |
888 | - Changed the Linux stack finding code to use /proc, after changing it | |
889 | to use HEURISTIC1. (Thanks to David Mossberger for pointing out the | |
890 | /proc hook.) | |
891 | - Added HP/UX incremental GC support and HP/UX 11 thread support. | |
892 | Thread support is currently still flakey. | |
893 | - Added basic Linux/IA64 support. | |
894 | - Integrated Anthony Green's PicoJava support. | |
895 | - Integrated Scott Ananian's StrongARM/NetBSD support. | |
896 | - Fixed some fairly serious performance bugs in the incremental | |
897 | collector. These have probably been there essentially forever. | |
898 | (Mark bits were sometimes set before scanning dirty pages. | |
899 | The reclaim phase unnecessarily dirtied full small object pages.) | |
900 | - Changed the reclaim phase to ignore nearly full pages to avoid | |
901 | touching them. | |
902 | - Limited GC_black_list_spacing to roughly the heap growth increment. | |
903 | - Changed full collection triggering heuristic to decrease full GC | |
904 | frequency by default, but to explicitly trigger full GCs during | |
905 | heap growth. This doesn't always improve things, but on average it's | |
906 | probably a win. | |
907 | - GC_debug_free(0, ...) failed. Thanks to Fergus Henderson for the | |
908 | bug report and fix. | |
909 | ||
910 | Since 5.0 alpha4 | |
911 | - GC_malloc_explicitly_typed and friends sometimes failed to | |
912 | initialize first word. | |
913 | - Added allocation routines and support in the marker for mark descriptors | |
914 | in a type structure referenced by the first word of an object. This was | |
915 | introduced to support gcj, but hopefully in a way that makes it | |
916 | generically useful. | |
917 | - Added GC_requested_heapsize, and inhibited collections in nonincremental | |
918 | mode if the actual used heap size is less than what was explicitly | |
919 | requested. | |
920 | - The Solaris pthreads version of GC_pthread_create didn't handle a NULL | |
921 | attribute pointer. Solaris thread support used the wrong default thread | |
922 | stack size. (Thanks to Melissa O'Neill for the patch.) | |
923 | - Changed PUSH_CONTENTS macro to no longer modify first parameter. | |
924 | This usually doesn't matter, but it was certainly an accident waiting | |
925 | to happen ... | |
926 | - Added GC_register_finalizer_no_order and friends to gc.h. They're | |
927 | needed by Java implementations. | |
928 | - Integrated a fix for a win32 deadlock resulting from clock() calling | |
929 | malloc. (Thanks to Chris Dodd.) | |
930 | - Integrated Hiroshi Kawashima's port to Linux/MIPS. This was designed | |
931 | for a handheld platform, and may or may not be sufficient for other | |
932 | machines. | |
933 | - Fixed a va_arg problem with the %c specifier in cordprnt.c. It appears | |
934 | that this was always broken, but recent versions of gcc are the first to | |
935 | report the (statically detectable) bug. | |
936 | - Added an attempt at a more general solution to dlopen races/deadlocks. | |
937 | GC_dlopen now temporarily disables collection. Still not ideal, but ... | |
938 | - Added -DUSE_I686_PREFETCH, -DUSE_3DNOW_PREFETCH, and support for IA64 | |
939 | prefetch instructions. May improve performance measurably, but I'm not | |
940 | sure the code will run correctly on processors that don't support the | |
941 | instruction. Won't build except with very recent gcc. | |
942 | - Added caching for header lookups in the marker. This seems to result | |
943 | in a barely measurable performance gain. Added support for interleaved | |
944 | lookups of two pointers, but unconfigured that since the performance | |
945 | gain is currently near zero, and it adds to code size. | |
946 | - Changed Linux DATA_START definition to check both data_start and | |
947 | __data_start, since nothing else seems to be portable. | |
948 | - Added -DUSE_LD_WRAP to optionally take advantage of the GNU ld function | |
949 | wrapping mechanism. Probably currently useful only on Linux. | |
950 | - Moved some variables for the scratch allocator into GC_arrays, on | |
951 | Martin Hirzel's suggestion. | |
952 | - Fixed a win32 threads bug that caused the collector to not look for | |
953 | interior pointers from one of the thread stacks without | |
954 | ALL_INTERIOR_POINTERS. (Thanks to Jeff Sturm.) | |
955 | - Added Mingw32 support. (Thanks again to Jeff Sturm for the patch.) | |
956 | - Changed the alpha port to use the generic register scanning code instead | |
957 | of alpha_mach_dep.s. Alpha_mach_dep.s doesn't look for pointers in fp | |
958 | registers, but gcc sometimes spills pointers there. (Thanks to Manuel | |
959 | Serrano for helping me debug this by email.) Changed the IA64 code to | |
960 | do something similar for similar reasons. | |
961 | ||
962 | [5.0alpha5 doesn't really exist, but it may have escaped.] | |
963 | ||
964 | Since 5.0alpha6: | |
965 | - -DREDIRECT_MALLOC was broken in alpha6. Fixed. | |
966 | - Cleaned up gc_ccp.h slightly, thus also causing the HP C++ compiler to | |
967 | accept it. | |
968 | - Removed accidental reference to dbg_mlc.c, which caused dbg_mlc.o to be | |
969 | linked into every executable. | |
970 | - Added PREFETCH to bitmap marker. Changed it to use the header cache. | |
971 | - GC_push_marked sometimes pushed one object too many, resulting in a | |
972 | segmentation fault in GC_mark_from_mark_stack. This was probably an old | |
973 | bug. It finally showed up in gctest on win32. | |
974 | - Gc_priv.h erroneously #defined GC_incremental to be TRUE instead of FALSE | |
975 | when SMALL_CONFIG was defined. This was no doubt a major performance bug for | |
976 | the default win32 configuration. | |
977 | - Removed -DSMALL_CONFIG from NT_MAKEFILE. It seemed like an anchronism now | |
978 | that the average PC has 64MB or so. | |
979 | - Integrated Bryce McKinley's patches for linux threads and dynamic loading | |
980 | from the libgcj tree. Turned on dynamic loading support for Linux/PPC. | |
981 | - Changed the stack finding code to use environ on HP/UX. (Thanks | |
982 | to Gustavo Rodriguez-Rivera for the suggestion.) This should probably | |
983 | be done on other platforms, too. Since I can't test those, that'll | |
984 | wait until after 5.0. | |
985 | ||
986 | Since 5.0alpha7: | |
987 | - Fixed threadlibs.c for linux threads. -DUSE_LD_WRAP was broken and | |
988 | -ldl was omitted. Fixed Linux stack finding code to handle | |
989 | -DUSE_LD_WRAP correctly. | |
990 | - Added MSWIN32 exception handler around marker, so that the collector | |
991 | can recover from root segments that are unmapped during the collection. | |
992 | This caused occasional failures under Windows 98, and may also be | |
993 | an issue under Windows NT/2000. | |
994 | ||
995 | Since 5.0 | |
996 | - Fixed a gc.h header bug which showed up under Irix. (Thanks to | |
997 | Dan Sullivan.) | |
998 | - Fixed a typo in GC_double_descr in typd_mlc.c. | |
999 | This probably could result in objects described by array descriptors not | |
1000 | getting traced correctly. (Thanks to Ben Hutchings for pointing this out.) | |
1001 | - The block nearly full tests in reclaim.c were not correct for 64 bit | |
1002 | environments. This could result in unnecessary heap growth under unlikely | |
1003 | conditions. | |
1004 | ||
1005 | Since 5.1 | |
1006 | - dyn_load.c declared GC_scratch_last_end_ptr as an extern even if it | |
1007 | was defined as a macro. This prevented the collector from building on | |
1008 | Irix. | |
1009 | - We quietly assumed that indirect mark descriptors were never 0. | |
1010 | Our own typed allocation interface violated that. This could result | |
1011 | in segmentation faults in the marker with typed allocation. | |
1012 | - Fixed a _DUSE_MUNMAP bug in the heap block allocation code. | |
1013 | (Thanks to Ben Hutchings for the patch.) | |
1014 | - Taught the collector about VC++ handling array operator new. | |
1015 | (Thanks again to Ben Hutchings for the patch.) | |
1016 | - The two copies of gc_hdrs.h had diverged. Made one a link to the other | |
1017 | again. | |
1018 | ||
1019 | Since 5.2 (A few 5.2 patches are not in 6.0alpha1) | |
1020 | - Fixed _end declaration for OSF1. | |
1021 | - There were lots of spurious leak reports in leak detection mode, caused | |
1022 | by the fact that some pages were not being swept, and hence unmarked | |
1023 | objects weren't making it onto free lists. (This bug dated back to 5.0.) | |
1024 | - Fixed a typo in the liblinuxgc.so Makefile rule. | |
1025 | - Added the GetExitCodeThread to Win32 GC_stop_world to (mostly) work | |
1026 | around a Windows 95 GetOpenFileName problem. (Thanks to Jacob Navia.) | |
1027 | ||
1028 | Since 5.3 | |
1029 | - Fixed a typo that prevented compilation with -DUSE_3DNOW_PREFETCH. | |
1030 | (Thanks to Shawn Wagner for actually testing this.) | |
1031 | - Fixed GC_is_thread_stack in solaris_threads.c. It forgot to return a value | |
1032 | in the common case. I wonder why nobody noticed? | |
1033 | - Fixed another silly syntax problem in GC_double_descr. (Thanks to | |
1034 | Fergus Henderson for finding it.) | |
1035 | - Fixed a GC_gcj_malloc bug: It tended to release the allocator lock twice. | |
1036 | ||
1037 | Since 5.4 (A few 5.3 patches are not in 6.0alpha2) | |
1038 | - Added HP/PA prefetch support. | |
1039 | - Added -DDBG_HDRS_ALL and -DSHORT_DBG_HDRS to reduce the cost and improve | |
1040 | the reliability of generating pointer backtrace information, e.g. in | |
1041 | the Bigloo environment. | |
1042 | - Added parallel marking support (-DPARALLEL_MARK). This currently | |
1043 | works only under IA32 and IA64 Linux, but it shouldn't be hard to adapt | |
1044 | to other platforms. This is intended to be a lighter-weight (less | |
1045 | new code, probably not as scalable) solution than the work by Toshio Endo | |
1046 | et al, at the University of Tokyo. A number of their ideas were | |
1047 | reused, though the code wasn't, and the underlying data structure | |
1048 | is significantly different. In particular, we keep the global mark | |
1049 | stack as a single shared data structure, but most of the work is done | |
1050 | on smaller thread-local mark stacks. | |
1051 | - Changed GC_malloc_many to be cheaper, and to require less mutual exclusion | |
1052 | with -DPARALLEL_MARK. | |
1053 | - Added full support for thread local allocation under Linux | |
1054 | (-DTHREAD_LOCAL_ALLOC). This is a thin veneer on GC_malloc_many, and | |
1055 | should be easily portable to other platforms, especially those that | |
1056 | support pthreads. | |
1057 | - CLEAR_DOUBLE was not always getting invoked when it should have been. | |
1058 | - GC_gcj_malloc and friends used different out of memory handling than | |
1059 | everything else, probably because I forgot about one when I implemented | |
1060 | the other. They now both call GC_oom_fn(), not GC_oom_action(). | |
1061 | - Integrated Jakub Jelinek's fixes for Linux/SPARC. | |
1062 | - Moved GC_objfreelist, GC_aobjfreelist, and GC_words_allocd out of | |
1063 | GC_arrays, and separately registered the first two as excluded roots. | |
1064 | This makes code compiled with gc_inl.h less dependent on the | |
1065 | collector version. (It would be nice to remove the inclusion of | |
1066 | gc_priv.h by gc_inl.h completely, but we're not there yet. The | |
1067 | locking definitions in gc_priv.h are still referenced.) | |
1068 | This change was later coniditoned on SEPARATE_GLOBALS, which | |
1069 | is not defined by default, since it involves a performance hit. | |
1070 | - Register GC_obj_kinds separately as an excluded root region. The | |
1071 | attempt to register it with GC_arrays was usually failing. (This wasn't | |
1072 | serious, but seemed to generate some confusion.) | |
1073 | - Moved backptr.h to gc_backptr.h. | |
1074 | ||
1075 | Since 6.0alpha1 | |
1076 | - Added USE_MARK_BYTES to reduce the need for compare-and-swap on platforms | |
1077 | for which that's expensive. | |
1078 | - Fixed a locking bug ib GC_gcj_malloc and some locking assertion problems. | |
1079 | - Added a missing volatile to OR_WORD and renamed the parameter to | |
1080 | GC_compare_and_swap so it's not a C++ reserved word. (Thanks to | |
1081 | Toshio Endo for pointing out both of those.) | |
1082 | - Changed Linux dynamic library registration code to look at /proc/self/maps | |
1083 | instead of the rld data structures when REDIRECT_MALLOC is defined. | |
1084 | Otherwise some of the rld data data structures may be prematurely garbage | |
1085 | collected. (Thanks to Eric Benson for helping to track this down.) | |
1086 | - Fixed USE_LD_WRAP a bit more, so it should now work without threads. | |
1087 | - Renamed XXX_THREADS macros to GC_XXX_THREADS for namespace correctness. | |
1088 | Tomporarily added some backward compatibility definitions. Renamed | |
1089 | USE_LD_WRAP to GC_USE_LD_WRAP. | |
1090 | - Many MACOSX POWERPC changes, some additions to the gctest output, and | |
1091 | a few minor generic bug fixes. (Thanks to Dietmar Planitzer.) | |
1092 | ||
1093 | Since 6.0 alpha2 | |
1094 | - Fixed the /proc/self/maps code to not seek, since that apparently is not | |
1095 | reliable across all interesting kernels. | |
1096 | - Fixed some compilation problems in the absence of PARALLEL_MARK | |
1097 | (introduced in alpha2). | |
1098 | - Fixed an algorithmic problem with PARALLEL_MARK. If work needs to | |
1099 | be given back to the main mark "stack", the BOTTOM entries of the local | |
1100 | stack should be given away, not the top ones. This has substantial | |
1101 | performance impact, especially for > 2 processors, from what I can tell. | |
1102 | - Extracted gc_lock.h from gc_priv.h. This should eventually make it a | |
1103 | bit easier to avoid including gc_priv.h in clients. | |
1104 | - Moved all include files to include/ and removed duplicate links to the | |
1105 | same file. The old scheme was a bad idea because it was too easy to get the | |
1106 | copies out of sync, and many systems don't support hard links. | |
1107 | Unfortunately, it's likely that I broke some of the non-Unix Makefiles in | |
1108 | the process, although I tried to update them appropriately. | |
1109 | - Removed the partial support for a copied nursery. It's not clear that | |
1110 | this would be a tremendous win, since we don't consistently lose to | |
1111 | generational copying collectors. And it would significantly complicate | |
1112 | many things. May be reintroduced if/when it really turns out to win. | |
1113 | - Removed references to IRIX_JDK_THREADS, since I believe there never | |
1114 | were and never will be any clients. | |
1115 | - Added some code to linux_threads.c to possibly support HPUX threads | |
1116 | using the Linux code. Unfortunately, it doesn't work yet, and is | |
1117 | currently disabled. | |
1118 | - Added support under Linux/X86 for saving the call chain, both in (debug) | |
1119 | objects for client debugging, and in GC_arrays._last_stack for GC | |
1120 | debugging. This was previously supported only under Solaris. It is | |
1121 | not enabled by default under X86, since it requires that code be compiled | |
1122 | to explicitly dave frame pointers on the call stack. (With gcc this | |
1123 | currently happens by default, but is often turned off explicitly.) | |
1124 | To turn it on, define SAVE_CALL_CHAIN. | |
1125 | ||
1126 | Since 6.0 alpha3 | |
1127 | - Moved up the detection of mostly full blocks to the initiatiation of the | |
1128 | sweep phase. This eliminates some lock conention in the PARALLEL_MARK case, | |
1129 | as multiple threads try to look at mostly full blocks concurrently. | |
1130 | - Restored the code in GC_malloc_many that grabs a prefix of the global | |
1131 | free list. This avoids the case in which every GC_malloc_many call | |
1132 | tries and fails to allocate a new heap block, and the returns a single | |
1133 | object from the global free list. | |
1134 | - Some minor fixes in new_hblk.c. (Attempted to build free lists in order | |
1135 | of increasing addresses instead of decreasing addresses for cache performance | |
1136 | reasons. But this seems to be only a very minor gain with -DEAGER_SWEEP, | |
1137 | and a loss in other cases. So the change was backed out.) | |
1138 | - Fixed some of the documentation. (Thanks in large part to Fergus | |
1139 | Henderson.) | |
1140 | - Fixed the Linux USE_PROC_FOR_LIBRARIES code to deal with apps that perform | |
1141 | large numbers of mmaps. (Thanks to Eric Benson.) Also fixed that code to | |
1142 | deal with short reads. | |
1143 | - Added GC_get_total_bytes(). | |
1144 | - Fixed leak detection mode to avoid spurious messages under linuxthreads. | |
1145 | (This should also now be easy for the other supported threads packages. | |
1146 | But the code is tricky enough that I'm hesitant to do it without being able | |
1147 | to test. Everything allocated in the GC thread support itself should be | |
1148 | explicitly deallocated.) | |
1149 | - Made it possible (with luck) to redirect malloc to GC_local_malloc. | |
1150 | ||
1151 | Since 6.0 alpha4 | |
1152 | - Changed the definition of GC_pause in linux_threads.c to use a volatile | |
1153 | asm. Some versions of gcc apparently optimize away writes to local volatile | |
1154 | variables. This caused poor locking behaviour starting at about | |
1155 | 4 processors. | |
1156 | - Added GC_start_blocking(), GC_end_blocking() calls and wrapper for sleep | |
1157 | to linux_threads.c. | |
1158 | The first two calls could be used to generally avoid sending GC signals to | |
1159 | blocked threads, avoiding both premature wakeups and unnecessary overhead. | |
1160 | - Fixed a serious bug in thread-local allocation. At thread termination, | |
1161 | GC_free could get called on small integers. Changed the code for thread | |
1162 | termination to more efficiently return left-over free-lists. | |
1163 | - Integrated Kjetil Matheussen's BeOS support. | |
1164 | - Rearranged the directory structure to create the doc and tests | |
1165 | subdirectories. | |
1166 | - Sort of integrated Eric Benson's patch for OSF1. This provided basic | |
1167 | OSF1 thread support by suitably extending hpux_irix_threads.c. Based | |
1168 | on earlier email conversations with David Butenhof, I suspect that it | |
1169 | will be more reliable in the long run to base this on linux_threads.c | |
1170 | instead. Thus I attempted to patch up linux_threads.c based on Eric's code. | |
1171 | The result is almost certainly broken, but hopefully close enough that | |
1172 | someone with access to a machine can pick it up. | |
1173 | - Integrated lots of minor changes from the NetBSD distribution. (These | |
1174 | were supplied by David Brownlee. I'm not sure about the original | |
1175 | authors.) | |
1176 | - Hacked a bit more on the HP/UX thread-support in linux_threads.c. It | |
1177 | now appears to work in the absence of incremental collection. Renamed | |
1178 | hpux_irix_threads.c back to irix_threads.c, and removed the attempt to | |
1179 | support HPUX there. | |
1180 | - Changed gc.h to define _REENTRANT in cases in which it should already | |
1181 | have been defined. It is still safer to also define it on the command | |
1182 | line. | |
1183 | ||
1184 | Since 6.0alpha5: | |
1185 | - Changed the definition of DATASTART on ALPHA and IA64, where data_start | |
1186 | and __data_start are not defined by earlier versions of glibc. This might | |
1187 | need to be fixed on other platforms as well. | |
1188 | - Changed the way the stack base and backing store base are found on IA64. | |
1189 | This should now remain reliable on future kernels. But since it relies | |
1190 | on /proc, it will no longer work in the simulated NUE environment. | |
1191 | - Made the call to random() in dbg_mlc.c with -DKEEP_BACK_PTRS dependent | |
1192 | on the OS. On non-Unix systems, rand() should be used instead. Handled | |
1193 | small RAND_MAX. (Thanks to Peter Ross for pointing this out.) | |
1194 | - Fixed the cord make rules to create the cord subdirectory, if necessary. | |
1195 | (Thanks to Doug Moen.) | |
1196 | - Changed fo_object_size calculation in finalize.c. Turned finalization | |
1197 | of nonheap object into a no-op. Removed anachronism from GC_size() | |
1198 | implementation. | |
1199 | - Changed GC_push_dirty call in solaris_threads.c to GC_push_selected. | |
1200 | It was missed in a previous renaming. (Thanks to Vladimir Tsichevski | |
1201 | for pointing this out.) | |
1202 | - Arranged to not not mask SIGABRT in linux_threads.c. (Thanks to Bryce | |
1203 | McKinlay.) | |
1204 | - Added GC_no_dls hook for applications that want to register their own | |
1205 | roots. | |
1206 | - Integrated Kjetil Matheussen's Amiga changes. | |
1207 | - Added FREEBSD_STACKBOTTOM. Changed the X86/FreeBSD port to use it. | |
1208 | (Thanks to Matthew Flatt.) | |
1209 | - Added pthread_detach interception for platforms supported by linux_threads.c | |
1210 | and irix_threads.c. Should also be added for Solaris? | |
1211 | - Changed the USE_MMAP code to check for the case in which we got the | |
1212 | high end of the address space, i.e. mem_ptr + mem_sz == 0. It appears | |
1213 | that this can happen under Solaris 7. It seems to be allowed by what | |
1214 | I would claim is an oversight in the mmap specification. (Thanks to Toshio | |
1215 | Endo for pointing out the problem.) | |
1216 | - Cleanup of linux_threads.c. Some code was originally cloned from | |
1217 | irix_threads.c and now unnecessary. Some comments were obviously wrong. | |
1218 | - (Mostly) fixed a longstanding problem with setting of dirty bits from | |
1219 | a signal handler. In the presence of threads, dirty bits could get lost, | |
1220 | since the etting of a bit in the bit vector was not atomic with respect | |
1221 | to other updates. The fix is 100% correct only for platforms for which | |
1222 | GC_test_and_set is defined. The goal is to make that all platforms with | |
1223 | thread support. Matters only if incremental GC and threads are both | |
1224 | enabled. | |
1225 | - made GC_all_interior_pointers (a.k.a. ALL_INTERIOR_POINTERS) an | |
1226 | initialization time, instead of build-time option. This is a | |
1227 | nontrivial, high risk change. It should slow down the code measurably | |
1228 | only if MERGE_SIZES is not defined, which is a very nonstandard | |
1229 | configuration. | |
1230 | - Added doc/README.environment, and implemented what it describes. This | |
1231 | allows a number of additional configuration options to be set through | |
1232 | the environment. It documents a few previously undocumented options. | |
1233 | - Integrated Eric Benson's leak testing improvements. | |
1234 | - Removed the option to throw away the beginning of each page (DISCARD_WORDS). | |
1235 | This became less and less useful as processors enforce stricter alignment. | |
1236 | And it hadn't been tested in ages, and was thus probably broken anyway. | |
1237 | ||
1238 | Since 6.0alpha6: | |
1239 | - Added GC_finalizer_notifier. Fixed GC_finalize_on_demand. (The variable | |
1240 | actually wasn't being tested at the right points. The build-time flag | |
1241 | was.) | |
1242 | - Added Tom Tromey's S390 Linux patch. | |
1243 | - Added code to push GC_finalize_now in GC_push_finalizer_structures. | |
1244 | (Thanks to Matthew Flatt.) | |
1245 | - Added GC_push_gc_structures() to push all GC internal roots. | |
1246 | - Integrated some FreeBSD changes from Matthew Flatt. | |
1247 | - It looks like USRSTACK is not always correctly defined under Solaris. | |
1248 | Hacked gcconfig.h to attempt to work around the problem. The result | |
1249 | is not well tested. (Thanks again to Matthew Flatt for pointing this | |
1250 | out. The gross hack is mine. - HB) | |
1251 | - Added Ji-Yong Chung's win32 threads and C++ fixes. | |
1252 | - Arranged for hpux_test_and_clear.s to no longer be needed or built. | |
1253 | It was causing build problems with gas, and it's not clear this is | |
1254 | better than the pthreads alternative on this platform. | |
1255 | - Some MINGW32 fixes from Hubert Garavel. | |
1256 | - Added Initial Hitachi SH4 port from Kaz Kojima. | |
1257 | - Ported thread-local allocation and parallel mark code to HP/UX on PA_RISC. | |
1258 | - Made include/gc_mark.h more public and separated out the really private | |
1259 | pieces. This is probably still not quite sufficient for clients that | |
1260 | want to supply their own kind of type information. But it's a start. | |
1261 | This involved lots of identifier renaming to make it namespace clean. | |
1262 | - Added GC_dont_precollect for clients that need complete control over | |
1263 | the root set. | |
1264 | - GC_is_visible didn't do the right thing with gcj objects. (Not that | |
1265 | many people are likely to care, but ...) | |
1266 | - Don't redefine read with GC_USE_LD_WRAP. | |
1267 | - Initial port to LINUX/HP_PA. Incremental collection and threads are not | |
1268 | yet supported. (Incremental collection should work if you have the | |
1269 | right kernel. Threads may work with a sufficiently patched pthread | |
1270 | library.) | |
1271 | - Changed gcconfig.h to recognize __i386__ as an alternative to i386 in | |
1272 | many places. (Thanks to Benjamin Lerman.) | |
1273 | - Made win32_threads.c more tolerant of detaching a thread that it didn't | |
1274 | know about. (Thanks to Paul Nash.) | |
1275 | - Added Makefile.am and configure.in from gcc to the distribution, with | |
1276 | minimal changes. For the moment, those are just placeholders. In the | |
1277 | future, we're planning to switch to a GNU-style build environment for | |
1278 | Un*x-like systems, though the old Makefile will remain as a backup. | |
1279 | - Turned off STUBBORN_ALLOC by default, and added it back as a Makefile | |
1280 | option. | |
1281 | - Redistributed some functions between malloc.c and mallocx.c, so that | |
1282 | simple statically linked apps no longer pull in mallocx.o. | |
1283 | - Changed large object allocation to clear the first and last few words | |
1284 | of each block before releassing the lock. Otherwise the marker could see | |
1285 | objects with nonsensical type descriptors. | |
1286 | - Fixed a couple of subtle problems that could result in not recognizing | |
1287 | interior pointers from the stack. (I believe these were introduced | |
1288 | in 6.0alpha6.) | |
1289 | - GC_debug_free_inner called GC_free, which tried to reacquire the | |
1290 | allocator lock, and hence deadlocked. (DBG_HDRS_ALL probably never worked | |
1291 | with threads?) | |
1292 | - Fixed several problems with back traces. Accidental references to a free | |
1293 | list could cause the free list pointer to be overwritten by a back pointer. | |
1294 | There seemed to be some problems with the encoding of root and finalizer | |
1295 | references. | |
1296 | ||
1297 | Since 6.0alpha7: | |
1298 | - Changed GC_debug_malloc_replacement and GC_debug_realloc_replacement | |
1299 | so that they compile under Irix. (Thanks to Dave Love.) | |
1300 | - Updated powerpc_macosx_mach_dep.s so that it works if the collector | |
1301 | is in a dynamic library. (Thanks to Andrew Begel.) | |
1302 | - Transformed README.debugging into debugging.html, updating and | |
1303 | expanding it in the process. Added gcdescr.html and tree.html | |
1304 | from the web site to the GC distribution. | |
1305 | - Fixed several problems related to PRINT_BLACK_LIST. This involved | |
1306 | restructuring some of the marker macros. | |
1307 | - Fixed some problems with the sizing of objects with debug information. | |
1308 | Finalization was broken KEEP_BACK_PTRS or PRINT_BLACK_LIST. Reduced the | |
1309 | object size with SHORT_DEBUG_HDRS by another word. | |
1310 | - The "Needed to allocate blacklisted ..." warning had inadvertently | |
1311 | been turned off by default, due to a buggy test in allchblk.c. Turned | |
1312 | it back on. | |
1313 | - Removed the marker macros to deal with 2 pointers in interleaved fashion. | |
1314 | They were messy and the performance improvement seemed minimal. We'll | |
1315 | leave such scheduling issues to the compiler. | |
1316 | - Changed Linux/PowerPC test to also check for __powerpc__ in response | |
1317 | to a discussion on the gcc mailing list. | |
1318 | - On Matthew Flatt's suggestion removed the "static" from the jmp_buf | |
1319 | declaration in GC_generic_push_regs. This was causing problems in | |
1320 | systems that register all of their own roots. It looks far more correct | |
1321 | to me without the "static" anyway. | |
1322 | - Fixed several problems with thread local allocation of pointerfree or | |
1323 | typed objects. The collector was reclaiming thread-local free lists, since | |
1324 | it wasn't following the link fields. | |
1325 | - There was apparently a long-standing race condition related to multithreaded | |
1326 | incremental collection. A collection could be started and a thread stopped | |
1327 | between the memory unprotect system call and the setting of the | |
1328 | corresponding dirt bit. I believe this did not affect Solaris or PCR, which | |
1329 | use a different dirty-bit implementation. Fixed this by installing | |
1330 | signal handlers with sigaction instead of signal, and disabling the thread | |
1331 | suspend signal while in the write-protect handler. (It is unclear | |
1332 | whether this scenario ever actually occurred. I found it while tracking | |
1333 | down the following:) | |
1334 | - Incremental collection did not cooperate correctly with the PARALLEL_MARK | |
1335 | implementation of GC_malloc_many or the local_malloc primitves. It still | |
1336 | doesn't work well, but it shouldn't lose memory anymore. | |
1337 | - Integrated some changes from the gcc source tree that I had previously | |
1338 | missed. (Thanks to Bryce McKinley for the reminder/diff.) | |
1339 | - Added Makefile.direct as a copy of the default Makefile, which would | |
1340 | normally be overwritten if configure is run. | |
1341 | - Changed the gc.tar target in Makefile.direct to embed the version number | |
1342 | in the gc directory name. This will affect future tar file distributions. | |
1343 | - Changed the Irix dynamic library finding code to no longer try to | |
1344 | eliminate writable text segments under Irix6.x, since that is probably no | |
1345 | longer necessary, and can apparently be unsafe on occasion. (Thanks to | |
1346 | Shiro Kawai for pointing this out.) | |
1347 | - GC_cleanup with GC_DEBUG enabled passed a real object base address to | |
1348 | GC_debug_register_finalizer_ignore_self, which expected a pointer past the | |
1349 | debug header. Call GC_register_finalizer_ignore_self instead, even with | |
1350 | debugging enabled. (Thanks to Jean-Daniel Fekete for catching this.) | |
1351 | - The collector didn't build with call chain saving enabled but NARGS=0. | |
1352 | (Thanks to Maarten Thibaut.) | |
1353 | - Fixed up the GNU-style build files enough so that they work in some | |
1354 | obvious cases. | |
1355 | - Added initial port to Digital Mars compiler for win32. (Thanks to Walter | |
1356 | Bright.) | |
1357 | ||
1358 | Since 6.0alpha8: | |
1359 | - added README.macros. | |
1360 | - Made gc.mak a symbolic link to work around winzip's tendency to ignore | |
1361 | hard links. | |
1362 | - Simplified the setting of NEED_FIND_LIMIT in os_dep.c, possibly breaking | |
1363 | it on untested platforms. | |
1364 | - Integrated initial GNU HURD port. (Thanks to Chris Lingard and Igor | |
1365 | Khavkine.) | |
1366 | - A few more fixes for Digital Mars compiler (Walter Bright). | |
1367 | - Fixed gcc version recognition. Renamed OPERATOR_NEW_ARRAY to | |
1368 | GC_OPERATOR_NEW_ARRAY. Changed GC_OPERATOR_NEW_ARRAY to be the default. | |
1369 | It can be overridden with -DGC_NO_OPERATOR_NEW_ARRAY. (Thanks to | |
1370 | Cesar Eduardo Barros.) | |
1371 | - Changed the byte size to free-list mapping in thread local allocation | |
1372 | so that size 0 allocations are handled correctly. | |
1373 | - Fixed Linux/MIPS stackbottom for new toolchain. (Thanks to Ryan Murray.) | |
1374 | - Changed finalization registration to invoke GC_oom_fn when it runs out | |
1375 | of memory. | |
1376 | - Removed lvalue cast in finalize.c. This caused some debug configurations | |
1377 | not to build with some non-gcc compilers. | |
1378 | ||
1379 | Since 6.0alpha9: | |
1380 | - Two more bug fixes for KEEP_BACK_PTRS and DBG_HDRS_ALL. | |
1381 | - Fixed a stack clearing problem that resulted in SIGILL with a | |
1382 | misaligned stack pointer for multithreaded SPARC builds. | |
1383 | - Integrated another HURD patch (thanks to Igor Khavkine). | |
1384 | ||
1385 | Since 6.0: | |
1386 | - Non-debug, atomic allocations could result in bogus smashed object | |
1387 | reports with debugging on. (Thanks to Patrick Doyle for the small | |
1388 | test case.) | |
1389 | - Fixed GC_get_register_stack_base (Itanium only) to work around a glibc | |
1390 | 2.2.4 bug. | |
1391 | - Initial port to HP/UX on Itanium. Thread support and both 32 and 64 | |
1392 | bit ABIs appear to work. Parallel mark support doesn't yet, due to | |
1393 | some inline assembly code issues. Thread local allocation does appear | |
1394 | to work. | |
1395 | - ifdef'ed out glibc2.1/Itanium workaround. I suspect nobody is using | |
1396 | that combination anymore. | |
1397 | - Added a patch to make new_gc_alloc.h usable with gcc3.0. (Thanks to | |
1398 | Dimitris Vyzovitis for the patch.) | |
1399 | - Debugged 64-bit support on HP/UX PA-RISC. | |
1400 | - Turned on dynamic loading support for FreeBSD/ELF. (Thanks to Peter | |
1401 | Housel.) | |
1402 | - Unregistering of finalizers with debugging allocation was broken. | |
1403 | (Thanks to Jani Kajala for the test case.) | |
1404 | - Old finalizers were not returned correctly from GC_debug_register_finalizer. | |
1405 | - Disabled MPROTECT_VDB for Linux/M68K based on a report that it doesn't work. | |
1406 | - Cleaned up some statistics gathering code in reclaim.c (Thanks to Walter | |
1407 | Bright.) | |
1408 | - Added some support for OpenBSD/ELF/Linux. (Thanks to Suzuki Toshiya.) | |
1409 | - Added Jakub Jelinek's patch to use dl_iterate_phdr for dynamic library | |
1410 | traversal to dyn_load.c. Changed it to weakly reference dl_iterate_phdr, | |
1411 | so that the old code is stilll used with old versions of glibc. | |
1412 | - Cleaned up feature test macros for various threads packages and | |
1413 | integrated (partially functional) FreeBSD threads code from Loren Rittle. | |
1414 | It's likely that the cleanup broke something, since it touched lots of | |
1415 | code. It's also likelly that it fixed some unreported bugs in the | |
1416 | less common thread implementations, since some of the original code | |
1417 | didn't stand up to close scrutiny. Support for the next pthreads | |
1418 | implementation should be easier to add. | |
1419 | ||
1420 | Since 6.1alpha1: | |
1421 | - No longer wrap read by default in multithreaded applications. It was | |
1422 | pointed out on the libgcj list that this holds the allocation lock for | |
1423 | way too long if the read blocks. For now, reads into the heap are | |
1424 | broken with incremental collection. It's possible to turn this back on | |
1425 | if you make sure that read calls don't block (e.g. by calling select | |
1426 | first). | |
1427 | - Fix ifdef in Solaris_threads.h to refer to GC_SOLARIS_THREADS. | |
1428 | - Added check for environment variable GC_IGNORE_GCJ_INFO. | |
1429 | - Added printing of stop-the-world GC times if GC_PRINT_STATS environment | |
1430 | variable is set. | |
1431 | - The calloc definition in leak_detector.h was missing parentheses, and | |
1432 | realloc was missing a second argument to GC_REALLOC. | |
1433 | (Thanks to Elrond (elrond<at>samba-tng.org).) | |
1434 | - Added GC_PRINT_BACK_HEIGHT environment variable and associated | |
1435 | code, mostly in the new file backgraph.c. See doc/README.environment. | |
1436 | - Added -DUSE_GLOBAL_ALLOC to work around a Windows NT issue. (Thanks to | |
1437 | Jonathan Clark.) | |
1438 | - Integrated port to NEC EWS4800 (MIPS-based workstation, with somewhat | |
1439 | different address-space layout). This may help for other machines with | |
1440 | holes in the data segment. (Thanks to Hironori Sakamoto.) | |
1441 | - Changed the order in which GC_push_roots and friends push things onto | |
1442 | the mark stack. GC_push_all calls need to come first, since we can't | |
1443 | necessarily recovere if those overflow the mark stack. (Thanks to | |
1444 | Matthew Flatt for tracking down the problem.) | |
1445 | - Some minor cleanups to mostly support the Intel compiler on Linux/IA64. | |
1446 | ||
1447 | Since 6.1 alpha2: | |
1448 | - Minor cleanup on the gcconfig.h section for SPARC. | |
1449 | - Minor fix to support Intel compiler for I386/Linux. (Thanks to Sven | |
1450 | Hartrumpf.) | |
1451 | - Added SPARC V9 (64-bit) support. (Thanks to Jeff Sturm.) | |
1452 | - Restructured the way in which we determine whether or not to keep | |
1453 | call stacks for debug allocation. By default SAVE_CALL_COUNT is | |
1454 | now zero on all platforms. Added SAVE_CALL_NARGS parameters. | |
1455 | If possible, use execinfo.h to capture call stack. (This should | |
1456 | add support for a number of new platforms, though often at | |
1457 | considerable runtime expense.) | |
1458 | - Try to print symbolic information for call stacks. On Linux, we | |
1459 | do this with a combination of execinfo.h and running addr2line in | |
1460 | a separate process. This is both much more expensive and much more | |
1461 | useful. Amazingly, it seems to be fast enough for most purposes. | |
1462 | - Redefined strdup if -DREDIRECT_MALLOC is given. | |
1463 | - Changed incremental collector and MPROTECT_VDB implementation so that, | |
1464 | under favorable conditions, pointerfree objects are not protected. | |
1465 | Added GC_incremental_protection_needs() to determine ahead of time whether | |
1466 | pointerfree objects may be protected. Replaced GC_write_hint() with | |
1467 | GC_remove_protection(). | |
1468 | - Added test for GC_ENABLE_INCREMENTAL environment variable. | |
1469 | - Made GC_time_limit runtime configurable. Added GC_PAUSE_TIME_TARGET | |
1470 | environment variable. | |
1471 | - Eliminated GC_page_sz, a duplicate of GC_page_size. | |
1472 | - Caused the Solaris and Irix thread creation primitives to call | |
1473 | GC_init_inner(). | |
1474 | ||
1475 | Since 6.1alpha3: | |
1476 | - Fixed typo in sparc_mach_dep.S, preventing the 64-bit version from | |
1477 | building. Increased 64-bit heap size limit in test.c slightly, since | |
1478 | a functional SPARC collector seems to slightly exceed the old limits. | |
1479 | (Thanks again to Jeff Sturm.) | |
1480 | - Use NPRGREG in solaris_threads.c, thus printing all registers if things | |
1481 | go wrong. | |
1482 | - Added GC_MARKERS environment variable to allow use of a single marker | |
1483 | thread on an MP without confusing the lock implementation. | |
1484 | - Collect much less aggressively in incremental mode with GC_TIME_UNLIMITED. | |
1485 | This is really a purely generational mode, and we can afford to | |
1486 | postpone the collection until the heap is (nearly) full. | |
1487 | - Remove read() wrapper for MPROTECT_VDB. It was causing more harm than | |
1488 | good. It is often no longer needed if system calls avoid writing to | |
1489 | pointerfull heap objects. | |
1490 | - Fix MACOSX test in gcconfig.h. (Thanks to John Clements.) | |
1491 | - Change GC_test_and_set so that it consistently has one argument. | |
1492 | Add spaces to ::: in powerpc assembly code in gc_locks.h. | |
1493 | (Thanks to Ryan Murray.) | |
1494 | - Fixed a formatting error in dbg_mlc.c. Added prototype to GC_abort() | |
1495 | declaration. (Thanks to Michael Smith.) | |
1496 | - Removed "source" argument to GC_find_start(). Eliminate GC_FIND_START(). | |
1497 | - Added win32 recognition code in configure.in. Changed some of the | |
1498 | dllimport/export defines in gc.h. (Thanks to Adam Megacz.) | |
1499 | - GC_malloc_many didn't set hb_last_reclaimed when it called | |
1500 | GC_reclaim_generic. (I'm not sure this matters much, but ...) | |
1501 | - Allocating uncollectable objects with debug information sometimes | |
1502 | allocated objects that were one byte too small, since uncollectable | |
1503 | objects don't have the extra byte added at the end. (Thanks to | |
1504 | Wink Saville for pointing this out.) | |
1505 | - Added a bit more assertion checking to make sure that gcj objects | |
1506 | on free lists never have a nonzero second word. | |
1507 | - Replaced BCC_MAKEFILE with an up-to-date one. (Thanks to | |
1508 | Andre Leiradella.) | |
1509 | - Upgraded libtool, cinfigure.in and some related files to hopefully | |
1510 | support NetBSD/SPARC. (Thanks to Adrian Bunk.) Unfortunately, | |
1511 | libtool 1.4.2 seemed to be buggy due to missing quotes in several | |
1512 | "test" invocations. Fixed those in the ltmain.sh script. | |
1513 | - Some win32-specific patches, including the introduction of | |
1514 | GC_CreateThread. (Thanks to Adam Megacz.) | |
1515 | - Merged in gcj changes from Anthony Green to support embedded systems. | |
1516 | - Tried to consistently rename preprocessed assembly files with a capital | |
1517 | .S extension. | |
1518 | - Use alpha_mach_dep.S on ALPHA again. It doesn't really matter, but this | |
1519 | makes our distribution consistent with the gcc one, avoiding future merge | |
1520 | problems. | |
1521 | - Move GET_MEM definition into gcconfig.h. Include gcconfig.h slightly | |
1522 | later in gc_priv.h to avoid forward references to ptr_t. | |
1523 | - Add some testing of local allocation to test.c. | |
1524 | - Change definition of INVALID_QTID in specific.h. The -1 value was used | |
1525 | inconsistently, and too likely to collide with a valid stack address. | |
1526 | Some general clean-up of specific.[ch]. Added assertions. (Thanks | |
1527 | to Michael Smith for tracking down an intermittent bug to this | |
1528 | general area. I'm not sure it has been squashed yet, however.) | |
1529 | - On Pthread systems it was not safe to call GC_malloc() between fork() | |
1530 | and exec(). According to the applicable standards, it doesn't appear | |
1531 | to be safe to call malloc() or many other libc functions either, thus | |
1532 | it's not clear this is fixable. Added experimental support for | |
1533 | -DHANDLE_FORK in linux_threads.c which tries to support it. It may | |
1534 | succeed if libc does the right thing. I'm not sure whether it does. | |
1535 | (Thanks to Kenneth Schalk for pointing out this issue.) | |
1536 | - Documented thread local allocation primitives to require an | |
1537 | explicit GC_init call. GC_init_parallel is no longer declared to | |
1538 | be a constructor function, since that isn't portable and often | |
1539 | seems to lead to initialization order problems. | |
1540 | - Changed gc_cpp.cc and gc_cpp.h in one more attempt to make them | |
1541 | compatible with Visual C++ 6. (Thanks to Wink Saville for the | |
1542 | patch.) | |
1543 | - Some more patches for Linux on HP PA-RISC. | |
1544 | - Added include/gc_allocator.h. It implements (hopefully) standard | |
1545 | conforming (as opposed to SGI-style) allocators that allocate | |
1546 | collectable (gc_allocator) or GC-traceable, but not collectable | |
1547 | (traceable_allocator) objects. This borrows heavily from libstc++, | |
1548 | which borrows heavily from the SGI implementation, this part of | |
1549 | which was written by Matt Austern. Changed test_cpp.cc to very | |
1550 | minimally test this. | |
1551 | - On Linux/X86, retry mmap with a different start argument. That should | |
1552 | allow the collector to use more (closer to 3GB) of the address space. | |
1553 | (Thanks to Jeffrey Mark Siskind for tracking this down.) | |
1554 | - Force 64 bit alignment with GCJ support. (Reflects Bryce McKinley's | |
1555 | patch to the gcc tree.) | |
1556 | - Refined the choice of sa_handler vs. sa_sigaction in GC_dirty_init | |
1557 | to accomodate some glibc5 systems. (Thanks to Dan Fandrich for the | |
1558 | patch.) | |
1559 | - Compensated for the fact that current versions of glibc set | |
1560 | __libc_stack_end incorrectly on Linux/IA64 while initialization code | |
1561 | is running. This could cause the collector to miss 16 bytes of | |
1562 | the memory stack if GC_malloc or friends where called before main(). | |
1563 | - Mostly integrated Takis Psarogiannakopoulos' port to DG/UX Inix 86. | |
1564 | This will probably take another iteration to work, since his | |
1565 | patch conflicted with the libtool upgrade. | |
1566 | - Added README.arm.cross containing some information about cross- | |
1567 | compiling to an ARM processor from Margaret Fleck. | |
1568 | ||
1569 | Since 6.1alpha4: | |
1570 | - Added GC_finalizer_mem_freed, and changed some of the code that | |
1571 | decided on heap expansion to look at it. Memory explicitly | |
1572 | deallocated by finalizers essentially needs to be counted as reclaimed | |
1573 | by the GC. Otherwise there are cases in which the heap can grow | |
1574 | unboundedly. (Thanks to Mark Reichert for the test case.) | |
1575 | - Integrated Adam Megacz patches to not scan dynamic libraries if | |
1576 | we are compiling with gcc on win32. Otherwise we need structured | |
1577 | exception handling to deal with asynchronously unmapped root | |
1578 | segments, and gcc doesn't directly support that. | |
1579 | - Integrated Anthony Green's patch to support Wine. | |
1580 | - GC_OPERATOR_NEW_ARRAY was misspelled OPERATOR_NEW_ARRAY in several | |
1581 | places, including gc_cpp.cc. (Thanks to Wink Saville for pointing | |
1582 | this out.) | |
1583 | - Integrated Loren James Rittle's Alpha FreeBSD patches. In | |
1584 | response to Richard Henderson's suggestion, these also | |
1585 | changed the declarations of symbols like _end on many platforms to | |
1586 | that they wouldn't mistakenly be declared as short data symbols. | |
1587 | - Integrated changes from the Debian distribution. (Thanks to Ryan Murray | |
1588 | for pointing these out.) Fix C++ comments in POWERPC port. Add ARM32 | |
1589 | incremental GC support. Get rid of USE_GENERIC_PUSH_REGS for alpha/Linux, | |
1590 | this time for real. Use va_copy to get rid of cord printf problems | |
1591 | (finally). | |
1592 | - Close file descriptor used to count cpus. Thanks to Jeff Sturm for | |
1593 | pointing out the omission. | |
1594 | - Don't just drop gcj free lists in GC_start_reclaim, since that can | |
1595 | eventually cause the marker to see a bogus mark descriptor in the | |
1596 | dropped objects. The usual symptom was a very intermittent segmentation | |
1597 | fault in the marker. This mattered only if one of the GC_gcj_malloc | |
1598 | variants was used. (Thanks to Michael Smith, Jeff Sturm, Bryce | |
1599 | McKinley and Tom Tromey for helping to track this down.) | |
1600 | - Fixed Linux and Solaris/64 SPARC configuration. (Thanks to David Miller, | |
1601 | Jeff Sturm, Tom Tromey, and Christian Joensson.) | |
1602 | - Fixed a typo in strdup definition. (Thanks to Gerard A Allan.) | |
1603 | - Changed Makefile.direct to invoke $(CC) to assemble alpha_mach_dep.S. | |
1604 | This is needed on Linux. I'm not sure whether it's better or worse | |
1605 | on Tru64. | |
1606 | - Changed gc_cpp.h once more to declare operator new and friends only in | |
1607 | a Microsoft environment. This may need further fine tuning. (Thanks to | |
1608 | Johannes Schmidt for pointing out that the older code breaks on gcc3.0.4.) | |
1609 | - Don't ever override strdup if it's already macro defined. (Thanks to | |
1610 | Adnan Ali for pointing out the problem.) | |
1611 | - Changed gc_cpp.h yet again to also overload placement new. Due to the | |
1612 | C++ overloading rules, the other overloaded new operations otherwise hide | |
1613 | placement new, which causes many STL uses to break. (Thanks to Reza | |
1614 | Shahidi for reporting this, and to Matt Austern for proposing a fix.) | |
1615 | - Integrated cygwin pthreads support from Dan Bonachea. | |
1616 | - Turn on DYNAMIC_LOADING for NetBSD. (Thanks to Krister Walfridsson.) | |
1617 | - Changed printing code to print more complete GC times. | |
1618 | - Applied Mark Mitchell's Irix patch to correct some bitrot. | |
1619 | - Clarified which object-printing routines in dbg_mlc.c should hold | |
1620 | the allocation lock. Restructured the code to allow reasonable object | |
1621 | printing with -DREDIRECT_MALLOC. | |
1622 | - Fix the Linux mmap code to always start with 0x1000 as the initial hint. | |
1623 | Minor patches for 64-bit AIX, particularly to STACKBOTTOM. | |
1624 | (Thanks again to Jeffrey Mark Siskind.) | |
1625 | - Renamed "SUSPENDED" flag for Solaris threads support to avoid a conflict | |
1626 | with a system header. (Thanks to Philp Brown.) | |
1627 | - Cause win32_threads.c to handle an out of range stack pointer correctly, | |
1628 | though currently with a warning. (Thanks to Jonathan Clark for | |
1629 | observing that win32 applications may temporarily use the stack | |
1630 | pointer for other purposes, and suggesting a fix. Unfortunately, it's | |
1631 | not clear that there is a complete solution to this problem.) | |
1632 | ||
1633 | Since 6.1alpha5: | |
1634 | - Added GC_MAXIMUM_HEAP_SIZE environment variable. | |
1635 | - Fix configure.in for MIPS/LINUX. (Thanks to H.J. Lu.) | |
1636 | - Double page hash table size for -DLARGE_CONFIG. | |
1637 | - Integrated Bo Thorsen's X86-64 support. | |
1638 | - STACKBOTTOM definition for LINUX/MIPS was partially changed back. | |
1639 | (Thanks to H.J. Lu and Hiroshi Kawashima for resolving this.) | |
1640 | - Replaced all occurrences of LINUX_DATA_START in gcconfig.h with | |
1641 | SEARCH_FOR_DATA_START. It doesn't hurt to falll back to a search. | |
1642 | And __data_start doesn't seem to get defined correctly of the GC | |
1643 | library is loaded with LD_PRELOAD, e.g. for leak detection. | |
1644 | - If the GC_find_leak environment variable is set, do a | |
1645 | atexit(GC_gcollect) to give us at least one chance to detect leaks. | |
1646 | This may report some very benign leaks, but ... | |
1647 | - Addeded REDIRECT_FREE. It's necessary if we want leak detection with | |
1648 | LD_PRELOAD. | |
1649 | - Defer printing of leaked objects, as for smashed objects. | |
1650 | - Fixed process and descriptor leak in GC_print_callers. Try for | |
1651 | line number even if we got function name.) | |
1652 | - Ported parallel GC support and thread local allocation to Alpha. | |
1653 | Not yet well-tested. | |
1654 | - Added GC_DUMP_REGULARLY and added finalization statistics to GC_dump(). | |
1655 | - Fixed Makefile.am to mention alpha_mach_dep.S instead of the defunct | |
1656 | alpha_mach_dep.s. (Thanks to Fergus Henderson.) | |
1657 | - Incorporated a change to new_gc_alloc.h, suggested by Johannes Schmidt, | |
1658 | which should make it work with gcc3.1. (I would still like to encourage | |
1659 | use of gc_allocator.h instead.) | |
1660 | - Use alpha_mach_dep.S only on Linux. (It's not clear that this is | |
1661 | optimal, but it otherwise didn't build on Tru64. Thanks to Fergus | |
1662 | Henderson.) | |
1663 | - Added ifdef to guard free() in os_dep.c. Otherwise we get a | |
1664 | compilation error on Irix. (Thanks to Dai Sato.) | |
1665 | - Added an experimental version of GC_memalign to mallocx.c. This can't | |
1666 | always work, since we don't handle alignment requests in the hblk-level | |
1667 | allocator, and we can't handle arbitrary pointer displacements unless | |
1668 | GC_all_interior_pointers is enabled. But it should work for alignment | |
1669 | requests up to HBLKSIZE. This is not yet documented in the standard | |
1670 | places. | |
1671 | - Finally debugged the OSF1/Tru64 thread support. This needs more testing, | |
1672 | since I needed to add a somewhat unconvincing workaround for signal | |
1673 | delivery issues that I don't yet completely understand. But it does | |
1674 | pass my tests, even in parallel GC mode. Incremental GC support is | |
1675 | disabled if thread support is enabled, due to the signal issues. | |
1676 | - Eliminated name-space-incorrect definition of _cdecl from gc_cpp.h. | |
1677 | - Added GC_debug_malloc_replacement and GC_debug_realloc_replacement | |
1678 | declarations to gc.h. On IA64, this is required for REDIRECT_MALLOC | |
1679 | to work correctly with these. | |
1680 | - Fixed Linux USE_PROC_FOR_LIBRARIES to work with a 64-bit /proc format. | |
1681 | ||
1682 | Since 6.1: | |
1683 | - Guard the test for GC_DUMP_REGULARLY in misc.c with | |
1684 | "#ifndef NO_DEBUGGING". Otherwise it fails to build with NO_DEBUGGING | |
1685 | defined. (Thanks to Manuel Serrano.) | |
1686 | - Message about retrying suspend signals was incorrectly generated even when | |
1687 | flag was not set. | |
1688 | - Cleaned up MACOSX/NEXT root registration code. There was apparently a | |
1689 | separate ifdef case in GC_register_data_segments() for no reason. | |
1690 | - Removed MPROTECT_VDB for MACOSX port, based on one negative report. | |
1691 | - Arrange for gc.h and friends to be correctly installed with GNU-style | |
1692 | "make install". | |
1693 | - Enable the GNU-style build facility include C++ support in the library | |
1694 | with --enable-cplusplus. (Thanks to Thomas Maier for some of the patch.) | |
1695 | - Mark from GC_thread_key in linux_threads.c, in case that's allocated | |
1696 | from the garbage collected heap, as it is with our own thread-specific | |
1697 | storage implementation. (Thanks to Jeff Sturm.) | |
1698 | - Mark all free list header blocks if they are heap allocated. This avoids | |
1699 | some unnecessary tracing. And it remains correct if we clear the | |
1700 | root set. (Thanks to Jeff Sturm for identifying the bug.) | |
1701 | - Improved S390/Linux support. Add S390/Linux 64-bit support. (Thanks | |
1702 | to Ulrich Weigand.) | |
1703 | - Corrected the spelling of GC_{M,C}ALLOC_EXPLICTLY_TYPED to | |
1704 | GC_{M,C}ALLOC_EXPLICITLY_TYPED in gc_typed.h. This is technically | |
1705 | an interface change. Based on the fact that nobody reported this, | |
1706 | I suspect/hope there were no clients. | |
1707 | - Cleaned up gc_typed.h so that (1) it adds an extern "C" declaration | |
1708 | when appropriate, (2) doesn't generate references to undefined internal | |
1709 | macros, and (3) allows easier manual construction of descriptors. | |
1710 | - Close the file descriptor used by GC_print_address_map(). | |
1711 | - Set the "close-on-exec" bit for various file descriptors maintained | |
1712 | for the collector's internal use. | |
1713 | - Added a hack to find memory segments owned by the system allocator | |
1714 | under win32. Based on my tests, this tends to eventually find all | |
1715 | segments, though it may take a while. There appear to be cleaner, | |
1716 | but slower solutions under NT/XP. But they rely on an API that's | |
1717 | unsupported under 9X. | |
1718 | - Changed Linux PowerPC stack finding to LINUX_STACKBOTTOM. (Thanks | |
1719 | to Akira Tagoh for pointing out that HEURISTIC1 doesn't work on | |
1720 | 64-bit kernels.) | |
1721 | - Added GC_set_free_space_divisor to avoid some Windows dll issues. | |
1722 | - Added FIXUP_POINTER, POINTER_SHIFT, POINTER_MASK to allow preprocessing | |
1723 | of candidate pointers for tagging, etc. | |
1724 | - Always lock around GC_notify_full_gc(). Simplified code for | |
1725 | invoking GC_notify_full_gc(). | |
1726 | - Changed the way DATASTART is defined on FreeBSD to be robust against | |
1727 | an unmapped page after etext. (Thanks to Hironori Sakamoto for | |
1728 | tracking down the intermittent failure.) | |
1729 | - Made GC_enable() and GC_disable() official. Deprecated direct update | |
1730 | of GC_dont_gc. Changed GC_gcollect to be a noop when garbage collection | |
1731 | is disabled. | |
1732 | - Call GC_register_dynamic_libraries before stopping the world on Linux, | |
1733 | in order to avoid a potential deadlock due to the dl_iterate_phdr lock. | |
1734 | - Introduced a more general mechanism for platform-dependent code to | |
1735 | decide whether the main data segment should be handled separately | |
1736 | from dynamic libraries, or registered by GC_register_dynamic_libraries. | |
1737 | The latter is more reliable and easier on Linux with dl_iterate_phdr. | |
1738 | ||
1739 | Since 6.2alpha1: | |
1740 | - Fixed the completely broken FreeBSD code in 6.2alpha1. (Thanks to | |
1741 | Hironori Sakamoto for the patch.) | |
1742 | - Changed IRIX reference in dbg_mlc.c to IRIX5. (Thanks to Marcus Herbert.) | |
1743 | - Attempted to work around the problems with .S filenames and the SGI | |
1744 | compiler. (Reported by several people. Untested.) | |
1745 | - Worked around an HP/UX make issue with the GNU-style build process. | |
1746 | - Fixed the --enable-cplusplus build machinery to allow builds without | |
1747 | a C++ compiler. (That was always the intent ...) | |
1748 | - Changed the debugging allocation macros to explicitly pass the return | |
1749 | address for Linux and XXXBSD on hardware for which we can't get stack | |
1750 | traces. Use __builtin_return_address(0) to generate it when possible. | |
1751 | Some of the configuration work was cleaned up (good) and moved to gc.h | |
1752 | (bad, but necessary). This should make leak detection more useful | |
1753 | on a number of platforms. (Thanks to Fabian Thylman for the suggestion.) | |
1754 | - Fixed compilation problems in dbg_mlc.c with GC_ADD_CALLER. | |
1755 | - Bumped revision number for dynamic library. | |
1756 | ||
1757 | Since 6.2alpha2: | |
1758 | - Don't include execinfo.h in os_dep.c when it's not needed, and may not exist. | |
1759 | ||
1760 | Since 6.2alpha3: | |
1761 | - Use LINUX_STACKBOTTOM for >= glibc2.2 on Linux/MIPS. (See Debian bug | |
1762 | # 177204) | |
1763 | - Integrated Jeff Sturm and Jesse Rosenstock's MACOSX threads patches. | |
1764 | - Integrated Grzegorz Jakacki's substantial GNU build patch. "Make dist" | |
1765 | should now work for the GNU build process. Documentation files | |
1766 | are installed under share/gc. | |
1767 | - Tweaked gc_cpp.h to again support the Borland compiler. (Thanks to | |
1768 | Rene Girard for pointing out the problems.) | |
1769 | - Updated BCC_MAKEFILE (thanks to Rene Girard). | |
1770 | - Added GC_ASSERT check for minimum thread stack size. | |
1771 | - Added --enable-gc-assertions. | |
1772 | - Added some web documentation to the distribution. Updated it in the | |
1773 | process. | |
1774 | - Separate gc_conf_macros.h from gc.h. | |
1775 | - Added generic GC_THREADS client-defined macro to set the appropriate | |
1776 | GC_XXX_THREADS internal macro. (gc_config_macros.h.) | |
1777 | - Add debugging versions of _ignore_off_page allocation primitves. | |
1778 | - Moved declarations of GC_make_closure and GC_debug_invoke_finalizer | |
1779 | from gc.h to gc_priv.h. | |
1780 | - Reset GC_fail_count even if only a small allocation succeeds. | |
1781 | - Integrated Brian Alliet's patch for dynamic library support on Darwin. | |
1782 | - gc_cpp.h's gc_cleanup destructor called GC_REGISTER_FINALIZER_IGNORE_SELF | |
1783 | when it should have called the lower case version, since it was | |
1784 | explicitly computing a base pointer. | |
1785 | ||
1786 | Since 6.2alpha4: | |
1787 | - GC_invoke_finalizers could, under rare conditions, set | |
1788 | GC_finalizer_mem_freed to an essentially random value. This could | |
1789 | possibly cause unbounded heap growth for long-running applications | |
1790 | under some conditions. (The bug was introduced in 6.1alpha5, and | |
1791 | is not in gcc3.3. Thanks to Ben Hutchings for finding it.) | |
1792 | - Attempted to sanitize the various DLL macros. GC_USE_DLL disappeared. | |
1793 | GC_DLL is used instead. All internal tests are now on GC_DLL. | |
1794 | README.macros is now more precise about the intended meaning. | |
1795 | - Include DllMain in the multithreaded win32 version only if the | |
1796 | collector is actually built as a dll. (Thanks to Mohan Embar for | |
1797 | a version of the patch.) | |
1798 | - Hide the cygwin threadAttach/Detach functions. They were violating our | |
1799 | namespace rules. | |
1800 | - Fixed an assertion in GC_check_heap_proc. Added GC_STATIC_ASSERT. | |
1801 | (Thanks again to Ben Hutchings.) | |
1802 | - Removed some obsolete definitions for Linux/PowerPC in gcconfig.h. | |
1803 | - CORD_cat was not rebalancing unbalanced trees in some cases, violating | |
1804 | a CORD invariant. Also tweaked the rebalancing rule for | |
1805 | CORD_cat_char_star. (Thanks to Alexandr Petrosian for the bug report | |
1806 | and patch.) | |
1807 | - Added hand-coded structured exception handling support to mark.c. | |
1808 | This should enable support of dynamic libraries under win32 with | |
1809 | gcc-compiled code. (Thanks to Ranjit Mathew for the patch.) | |
1810 | Turned on dynamic library scanning for win32/gcc. | |
1811 | - Removed some remnants of read wrapping. (Thanks to Kenneth Schalk.) | |
1812 | GC_USE_LD_WRAP ws probably broken in recent versions. | |
1813 | - The build could fail on some platforms since gcconfig.h could include | |
1814 | declarations mentioning ptr_t, which was not defined, e.g. when if_mach | |
1815 | was built. (Thanks to Yann Dirson for pointing this out.) Also | |
1816 | cleaned up tests for GC_PRIVATE_H in gcconfig.h a bit. | |
1817 | - The GC_LOOP_ON_ABORT environment variable interfered with incremental | |
1818 | collection, since the write fault handler was erroneously overridden. | |
1819 | Handlers are now set up in the correct order. | |
1820 | - It used to be possible to call GC_mark_thread_local_free_lists() while | |
1821 | the world was not stopped during an incremental GC. This was not safe. | |
1822 | Fortunately, it was also unnecessary. Added GC_world_stopped flag | |
1823 | to avoid it. (This caused occasional crashes in GC_set_fl_marks | |
1824 | with thread local allocation and incremental GC. This probably happened | |
1825 | primarily on old, slow multiprocessors.) | |
1826 | - Allowed overriding of MAX_THREADS in win32_threads.c from the build | |
1827 | command line. (Patch from Yannis Bres.) | |
1828 | - Taught the IA64/linux code to determine the register backing store base from | |
1829 | /proc/self/maps after checking the __libc symbol, but before guessing. | |
1830 | (__libc symbols are on the endangered list, and the guess is likely to not | |
1831 | always be right for 2.6 kernels.) Restructured the code to read and parse | |
1832 | /proc/self/maps so it only exists in one place (all platforms). | |
1833 | - The -DUSE_PROC_FOR_LIBRARIES code was broken on Linux. It claimed that it | |
1834 | also registered the main data segment, but didn't actually do so. (I don't | |
1835 | think anyone actually uses this configuration, but ...) | |
1836 | - Made another attempt to get --enablecplusplus to do the right thing. | |
1837 | Since there are unavoidable problems with C programs linking against a | |
1838 | dynamic library that includes C++ code, I separated out the c++ code into | |
1839 | libgccpp. | |
1840 | ||
1841 | Since 6.2alpha5: | |
1842 | - There was an extra underscore in the name of GC_save_registers_in_stack | |
1843 | for NetBSD/SPARC. (Thanks to Jaap Boender for the patch.) | |
1844 | - Integrated Brian Alliet's patch for Darwin. This restructured the | |
1845 | linuxthreads/pthreads support to separate generic pthreads support | |
1846 | from more the system-dependent thread-stopping code. I believe this | |
1847 | should make it easier to eliminate the code duplication between | |
1848 | pthreads platforms in the future. The patch included some other | |
1849 | code cleanups. | |
1850 | - Integrated Dan Bonachea's patch to support AIX threads. This required | |
1851 | substantial manual integration, mostly due to conflicts with other | |
1852 | recent threads changes. It may take another iteration to | |
1853 | get it to work. | |
1854 | - Removed HPUX/PA-RISC support from aix_irix_threads.c. It wasn't used | |
1855 | anyway and it cluttered up the code. And anything we can do to migrate | |
1856 | towards generic pthreads support is a good thing. | |
1857 | - Added a more explicit test for tracing of function arguments to test.c. | |
1858 | (Thanks to Dan Grayson.) | |
1859 | - Added Akira Tagoh's PowerPC64 patch. | |
1860 | - Fixed some bit rot in the Cygwin port. (Thanks to Dan Bonachea for | |
1861 | pointing it out.) Gc.h now includes just windows.h, not winbase.h. | |
1862 | - Declared GC_save_regs_in_stack() in gc_priv.h. Remove other declarations. | |
1863 | - Changed --enable-cplusplus to use automake consitionals. The old way | |
1864 | confused libtool. "Make install" didn't work correctly for the old version. | |
1865 | Previously --enable-cplusplus was broken on cygwin. | |
1866 | - Changed the C version of GC_push_regs to fail at compile time if it is | |
1867 | generated with an empty body. This seems to have been the cause of one | |
1868 | or two subtle failures on unusual platforms. Those failures should | |
1869 | now occur at build time and be easily fixable. | |
1870 | ||
1871 | Since 6.2alpha6: | |
1872 | - Integrated a second round of Irix/AIX patches from Dan Bonachea. | |
1873 | Renamed mips_sgi_mach_dep.S back to mips_sgi_mach_dep.s, since it requires | |
1874 | the Irix assembler to do the C preprocessing; gcc -E doesn't work. | |
1875 | - Fixed Makefile.direct for DARWIN. (Thanks to Manuel Serrano.) | |
1876 | - There was a race between GC_pthread_detach and thread exit that could | |
1877 | result in a thread structure being deallocated by GC_pthread_detach | |
1878 | eventhough it was still needed by the thread exit code. (Thanks to | |
1879 | Dick Porter for the small test case that allowed this to be debugged.) | |
1880 | - Fixed version parsing for non-alpha versions in acinclude.m4 and | |
1881 | version checking in version.h. | |
1882 | ||
1883 | Since 6.2: | |
1884 | - Integrated some NetBSD patches forwarded to me by Marc Recht. These | |
1885 | were already in the NetBSD package. | |
1886 | - GC_pthread_create waited for the semaphore even if pthread_create failed. | |
1887 | Thanks to Dick Porter for the pthread_support.c patch. Applied the | |
1888 | analogous fix for aix_irix_threads.c. | |
1889 | - Added Rainer Orth's Tru64 fixes. | |
1890 | - The check for exceeding the thread table size in win32 threadDetach | |
1891 | was incorrect. (Thanks to Alexandr Petrosian for the patch.) | |
1892 | - Applied Andrew Begel's patch to correct some reentrancy issues | |
1893 | with dynamic loading on Darwin. | |
1894 | - GC_CreateThread() was neglecting to duplicate the thread handle in | |
1895 | the table. (Thanks to Tum Nguyen for the patch.) | |
1896 | - Pass +ESdbgasm only on PA-RISC machines with vendor compiler. | |
1897 | (Thanks to Roger Sayle for the patch.) | |
1898 | - Applied more AIX threads patches from Scott Ananian. | |
1899 | ||
1900 | Since 6.3alpha1: | |
1901 | - Reenabled I_HOLD_LOCK assertion in aix_irix_threads.h. | |
1902 | - Put back the WINABI qualifier for GC_CreateThread. (Thanks to | |
1903 | Danny Smith for the patch. 6.3alpha1 had the qualifier in one place | |
1904 | but not elsewhere, which was clearly wrong.) | |
1905 | - Sometimes explicitly define __private_extern__ before DARWIN dyld.h | |
1906 | include. (Thanks to Andreas Tobker for postting the patch.) | |
1907 | - Included signal.h from pthread_support.c. Removed GC_looping_handler, | |
1908 | which was dead code. | |
1909 | - GC_find_start was misdeclared by gc_pmark.h if PRINT_BLACK_LIST was | |
1910 | defined. (Thanks to Glauco Masotti for testing and reporting this.) | |
1911 | Changed GC_find_start to never just return 0. According to its | |
1912 | comment it doesn't, and it's unclear that's correct. | |
1913 | - GC_alloc_large had several largely compensating bugs in the | |
1914 | computation of GC_words_wasted. (It was confused about bytes vs. | |
1915 | words in two places.) | |
1916 | - Integrated Slava Sysoltev's patch to support more recent versions of | |
1917 | the Intel compiler on IA64/Linux. | |
1918 | - Changed win32 spinlock initialization to conditionally set a spin count. | |
1919 | (Emmanual Stumpf pointed out that enabling this makes a large performance | |
1920 | difference on win32 multiprocessors.) Also cleaned up the win32 spinlock | |
1921 | initialization code a bit. | |
1922 | - Fixed thread support for HP/UX/IA64. The register backing store base for | |
1923 | the main thread was sometimes not set correctly. (Thanks to Laurent | |
1924 | Morichetti.) | |
1925 | - Added -DEMPTY_GETENV_RESULTS flag to work around Wine problem. | |
1926 | - Declare GC_stack_alloc and GC_stack_free in solaris_threads.h to | |
1927 | avoid 64-bit size mismatches. (Thanks to Bernie Solomon.) | |
1928 | - Fixed GC_generic_push_regs to avoid a potential and very unfortunate | |
1929 | tail call optimization. This could lead to prematurely reclaimed | |
1930 | objects on configurations that used the generic routine and the new | |
1931 | build infrastructure (which potentially optimizes mach_dep.c). | |
1932 | This was a serious bug, but it's unclear whether it has resulted in | |
1933 | any real failures. | |
1934 | - Fixed CORD_str to deal with signed characters. (Thanks to Alexandr | |
1935 | Petrosian for noticing the problem and supplying the patch.) | |
1936 | - Merged a couple of NOSYS/ECOS tests into os_dep.c from gcj. (Thanks | |
1937 | to Anthony Green.) | |
1938 | - Partially merged a win32 patch from Ben Hutchings, and substantially | |
1939 | revised other parts of win32_threads.c. It had several problems. | |
1940 | Under MinGW with a statically linked library, the main thread was | |
1941 | not registered. Cygwin detached threads leaked thread descriptors. | |
1942 | There were several race conditions. For now, unfortunately the | |
1943 | static threads limit remains, though we increased it, and made table | |
1944 | traversal cost depend on the actual thread count. | |
1945 | There is also still some code duplication with pthread_support.c. | |
1946 | (Thread descriptors did become much smaller, since Ben Hutchings | |
1947 | removed the thread context from them.) | |
1948 | - Integrated a Solaris configure.in patch from Rainer Orth. | |
1949 | - Added GC_IGNORE_FB and associated warning to very partially address | |
1950 | the issue of the collector treating a mapped frame buffer as part | |
1951 | of the root set. (Thanks to David Peroutka for providing some | |
1952 | insight. More would be helpful. Is there anything that can be used | |
1953 | to at least partially identify such memory segments?) | |
1954 | ||
1955 | Since 6.3alpha2: | |
1956 | - Removed -DSMALL_CONFIG from BCC_MAKEFILE. | |
1957 | - Changed macros to test for an ARM processor (Patch from Richard Earnshaw.) | |
1958 | - Mostly applied a DJGPP patch from Doug Kaufman. Especially Makefile.dj | |
1959 | had suffered from serious bit rot. | |
1960 | - Rewrote GC_apply_to_maps, eliminating an off-by-one subscript error, | |
1961 | and a call to alloca (for lcc compatibility). | |
1962 | - Changed USE_MUNMAP behavior on posixy platforms to immediately remap | |
1963 | the memory with PROT_NONE instead of unmapping it. The latter risks | |
1964 | an intervening mmap grabbing the address space out from underneath us. | |
1965 | Updated this code to reflect a cleaner patch from Ulrich Drepper. | |
1966 | - Replaced _T with _Tp in new_gc_alloc.h to avoid a MACOS X conflict. | |
1967 | (Patch from Andrew Begel.) | |
1968 | - Dynamically choose whether or not lock should spin on win32. (Thanks | |
1969 | to Maurizio Vairani for the patch.) This may be a significant performance | |
1970 | improvement for win32. | |
1971 | - Fix Makefile.direct to actually include NT_STATIC_THREADS_MAKEFILE | |
1972 | in the distribution. (Again thanks to Maurizio Vairani.) | |
1973 | - Maybe_install_looping_handler() was accidentally exported, violating | |
1974 | our name space convention. | |
1975 | - Made os_dep.c use sigsetjmp and SA_NODEFER for NetBSD. (Thanks to | |
1976 | Christian Limpach.) (I generalized the patch to use sigsetjmp on all | |
1977 | UNIX_LIKE platforms, admittedly a slightly risky move. But it may avoid | |
1978 | similar problems on some other platforms. I also cleaned up the defn | |
1979 | of UNIX_LIKE a bit. - Hans) | |
1980 | - Integrated Andrew Begel's Darwin threads patch, adjusted according to | |
1981 | some of Fergus Hendersons's comments. (Patch didn't apply cleanly, | |
1982 | errors are possible.) | |
1983 | - Added another test or two for the Intel 8.0 compiler to avoid | |
1984 | confusing it with gcc. The single-threaded collector should now build | |
1985 | with icc, at least on ia64. | |
1986 | ||
1987 | Since 6.3alpha3: | |
1988 | - USE_MMAP was broken by confusion in the code dealing with USE_MMAP_ANON. | |
1989 | (This was pointed out, and fixes were suggested by several other people.) | |
1990 | - Darwin supprt was broken in alpha3 as a result of my misintegration of | |
1991 | Andrew Begel's patches. Fixed with another patch from Andrew Begel. | |
1992 | - A new sanity check in pthread_stop_world.c:GC_push_all_stacks() was | |
1993 | overly aggressive. We may collect from an unregistered thread during | |
1994 | thread creation. Fixed by explicitly checking for that case. (Added | |
1995 | GC_in_thread_creation.) | |
1996 | ||
1997 | Since 6.3alpha4: | |
1998 | - Fix & vs && typo in GC_generic_malloc and | |
1999 | GC_generic_malloc_ignore_off_page. (Propagated from the gcc tree.) | |
2000 | - Removed SA_NODEFER hack from NetBSD and Solaris write-protect handler. | |
2001 | (According to Christian Limpach, the NetBSD problem is fixed. | |
2002 | Presumably so is the Solaris 2.3 problem.) | |
2003 | - Removed placement delete from gc_cpp.h for the SGI compiler. | |
2004 | (Thanks to Simon Gornall for the patch.) | |
2005 | - Changed semantics of the GC_IGNORE_FB environment variable, based | |
2006 | on experimentation by Nicolas Cannasse pointing out that the old | |
2007 | interpretation was useless. We still need help in identifying win32 | |
2008 | graphics memory mappings. The current "solution" is a hack. | |
2009 | - Removed "MAKEOVERRIDES =" from Makefile.am and thus Makefile.in. | |
2010 | It probably made more sense in the gcc context. | |
2011 | - Explicitly ensure that NEED_FIND_LIMIT is defined for {Open,Net}BSD/ELF. | |
2012 | - Replaced USE_HPUX_TLS macro by USE_COMPILER_TLS, since gcc often | |
2013 | supports the same extension on various platforms. | |
2014 | - Added some basic (completely untested) defines for win64, in support | |
2015 | of future work. | |
2016 | - Declared GC_jmp_buf in os_dep.s as JMP_BUF instead of jmp_buf, fixing | |
2017 | a memory overwrite bug on Solaris and perhaps other platforms. | |
2018 | - Added 0 != __libc_stack_end test to GC_linux_stack_base. (Thanks to Jakub | |
2019 | Jelinek, both for the patch, and for explaining the problem to me.) | |
2020 | Otherwise "prelink"ing could cause the collector to fail. | |
2021 | - Changed default thread local storage implementation to USE_PTHREAD_SPECIFIC | |
2022 | for HP/UX with gcc. The compiler-based implementation appears to work | |
2023 | only with the vendor compiler. | |
2024 | - Export GC_debug_header_size and GC_USR_PTR_FROM_BASE from gc_mark.h, | |
2025 | making client mark code cleaner and less dependent on GC version. | |
2026 | - Export several new procedures and GC_generic_malloc from gc_mark.h | |
2027 | to support user-defined kinds. Use the new procedures to replace existing | |
2028 | code in gcj_mlc.c and typd_mlc.c. | |
2029 | - Added support for GC_BACKTRACES. | |
2030 | - Fixed a remaining problem in CORD_str with signed characters. (Thanks | |
2031 | to Alexandr Petrosian for the patch.) | |
2032 | - Removed supposedly redundant, but very buggy, definitions of finalizer | |
2033 | macros from javaxfc.h. Fortunately this file probably has no users. | |
2034 | The correct declarations were already in gc.h. | |
2035 | - Also need to set GC_in_thread_creation while waiting for GC during | |
2036 | thread termination, since it is also possible to collect from an | |
2037 | unregistered thread in that case. | |
2038 | - Define NO_GETENV for Windows CE, since getenv doesn't appear to exist. | |
2039 | + some other minor WinCE fixes. (Thanks to Alain Novak.) | |
2040 | - Added GC_register_describe_type_fn. | |
2041 | - Arrange for debugging finalizer registration to ignore non-heap | |
2042 | registrations, since the regular version of the routine also behaves | |
2043 | that way. | |
2044 | - GC_gcj_malloc and friends need to check for finalizers waiting to be run. | |
2045 | One of the more obscure allocation routines with missing a LOCK() call. | |
2046 | - Fixed cvtres invocations in NT_MAKEFILE and NT_STATIC_THREADS_MAKEFILE | |
2047 | to work with VS.NET. | |
2048 | - Cleaned up GC_INIT calls in test. Updated gc.man to encourage GC_INIT | |
2049 | use in portable code. | |
2050 | - Taught the GC to use libunwind if --enable-full-debug is specified on | |
2051 | IA64 and libunwind is present. | |
2052 | - The USE_MUNMAP code could get confused about the age of a block and | |
2053 | prematurely unmap it. GC_unmap_old had a bug related to wrapping of | |
2054 | GC_gc_no. GC_freehblk and GC_merge_unmapped didn't maintain | |
2055 | hb_last_reclaimed reasonably when blocks were merged. The code was | |
2056 | fixed to reflect original intent, but that may not always be an | |
2057 | improvement. See todo list item. | |
2058 | ||
2059 | Since 6.3alpha5: | |
2060 | - Define USE_GENERIC_PUSH_REGS for NetBSD/M68K. | |
2061 | - Fixed the X86_64 PREFETCH macros to correctly handle ia32e (which uses | |
2062 | different prefetch instructions from AMD64). (Thanks to H.J. Lu.) | |
2063 | - GC_config_macros.h did not correctly define GC_WIN32_THREADS from | |
2064 | GC_THREADS. | |
2065 | - Added simple_example.html. | |
2066 | - Merged Andrew Gray's patch to correctly restore signal handlers on | |
2067 | FreeBSD. | |
2068 | - Merged a patch from Andreas Jaeger to deal with prefetch-related warnings | |
2069 | on x86-64. Added some other casts so that the PREFETCH macros | |
2070 | always get a ptr_t argument. Removed some casts inthe PREFETCH | |
2071 | implementations. | |
2072 | - At Jesse Jones suggestion: Added a header guard for gc_allocator.h | |
2073 | and changed GC_debug_free to clobber contents of deallocated object. | |
2074 | - The signal masking code in pthread_stop_world.c contained some errors. | |
2075 | In particular SIGSEGV was masked in the handler, in spite of the fact that | |
2076 | it wrote to the heap. This could lead to an uncaught SIGSEGV, which | |
2077 | apparently became much more likely in Linux 2.6. Also fixed some | |
2078 | typos, and reduced code duplication in the same area. | |
2079 | - Remove ltconfig, clean up configure messages for DGUX (thanks to | |
2080 | Adrian Bunk for the patches). | |
2081 | - Integrated NetBSD/OpenBSD patches from Marc Recht and Matthias Drochner. | |
2082 | ||
2083 | Since 6.3alpha6: | |
2084 | - Compile test_cpp.cc with CXXCOMPILE instead of COMPILE. | |
2085 | - Very large allocations could cause a collector hang. Correct | |
2086 | calculation of GC_collect_at_heapsize. | |
2087 | - GC_print_hblkfreelist printed some bogus results if USE_MUNMAP | |
2088 | was defined. | |
2089 | - Include gc_config_macros.h in threadlibs.c. | |
2090 | - Correct MacOSX thread stop code. (Thanks to Dick Porter.) | |
2091 | - SMALL_OBJ definition was off by one. This could cause crashes | |
2092 | at startup. (Thanks to Zoltan Varga for narrowing this down to | |
2093 | a trivial test case.) | |
2094 | - Integrate Paolo Molara's patch to deal with a race in the Darwin | |
2095 | thread stopping code. | |
2096 | - Changed X86_64 implementation to use SA_SIGINFO in the MPROTECT_VDB | |
2097 | implementation. The old approach appears to have been broken by | |
2098 | recent kernels. | |
2099 | - Added GC_ATTR_UNUSED to eliminate a warning in gc_allocator.h (Thanks | |
2100 | to Andrew Begel.) | |
2101 | - Fix GC_task_self declaration in os_dep.c. (Thanks to Andrew Pinski.) | |
2102 | - Increase INITIAL_BUF_SZ in os_dep.c for Solaris /proc reads. | |
2103 | ||
2104 | Since gc6.3: | |
2105 | - Merge gcconfig.h changes from gcc tree. | |
2106 | - Unconditionally include gc_priv.h in solaris_pthreads.c, win32_threads.h, | |
2107 | aix_irix_threads.c, and solaris_threads.c to get thread definitions. | |
2108 | - Start marker threads in GC_thr_init, so that they get started even | |
2109 | if no other threads are ever started. (Oddly enough, the parallel | |
2110 | collector worked correctly, though not well, with no helper threads.) | |
2111 | - Go ahead and split large blocks in GC_allochblk_nth if GC_dont_gc | |
2112 | is set. (Thanks to Alexander Petrossian.) | |
2113 | - GC_PRINT_BACK_HEIGHT would deadlock with thread support. | |
2114 | - Let in_progress_space in backgraph.s grow dynamically. | |
2115 | - Fix README.solaris2. The GC_thr_init() hack doesn't work anymore. | |
2116 | - Convert GC_finalizer_mem_freed to bytes in allchblk.c. | |
2117 | - Add missing declaration for GC_generic_malloc_words_small_inner. | |
2118 | Without it, s390x breaks. (Thanks to Ulrich Weigand.) | |
2119 | - Applied several MacOSX patches to support older tool chains. | |
2120 | (Thanks to Stefan Ring.) | |
2121 | - Bug fix for NetBSD/amd64. (Thanks to Marc Recht.) Add NetBSD/sh3 | |
2122 | support. (Thanks to Uchiyama Yasushi.) | |
2123 | - Fixed an uninitialized variable in cordprnt.c. (Thanks to gcc for | |
2124 | providing the warning.) | |
2125 | - Eliminated some, but not all, gcc -Wall warnings. | |
2126 | - Changed some old style casts to reinterpret_cast in new_gc_alloc.h. | |
2127 | (Thanks to Dan Grayson.) | |
2128 | - GC_extend_size_map shouldn't adjust for GC_all_interior_pointers if | |
2129 | GC_DONT_ADD_BYTE_AT_END is set. | |
2130 | - Changed some (long) casts to (word) in preparation for win64. | |
2131 | (Thanks to Peter Colson.) | |
2132 | - Changed "int stack_size" declaration in pthread_support.c to use | |
2133 | size_t. (Only mattered with GC_ASSERTIONS enabled.) | |
2134 | - Added CRIS (etrax) support. (Thanks to Simon Posnjak and | |
2135 | Hans-Peter Nilsson.) | |
2136 | - Removed GC_IGNORE_FB frame buffer recognition, and replaced | |
2137 | it with a check that the mapping type is MEM_IMAGE. | |
2138 | In theory, this should work much better, but it is a high | |
2139 | risk change for win32. (Thanks to Ashley Bone for the crucial | |
2140 | experimental data behind this, and to Rutger Ovidus for | |
2141 | some further experiments.) | |
2142 | - GC_allochblk_nth incremented GC_words_wasted by bytes rather than | |
2143 | words. | |
2144 | - Consider GC_words_wasted in GC_adj_words_allocd only if it is within | |
2145 | reason. (A hack to avoid some extremely unlikely scenarios in which | |
2146 | we manage to allocate only "wasted" space. 7.0 has a better fix.) | |
2147 | - Changed PowerPC GC_clear implementation to use lwsync instead of | |
2148 | eieio, since the documentation recommends against eieio, and | |
2149 | it seems to be incorrect if the preceding memory op is a load. | |
2150 | - Fixed print_block_list to print the correct kind number for | |
2151 | STUBBORN. (Thanks to Rutger Ovidus.) | |
2152 | - Have configure.in generate an error if it is asked to support | |
2153 | pthreads, but doesn't know how to. | |
2154 | - Added Kazuhiro Inaoka's patch for Renesas M32R support. | |
2155 | - Have the GNU build mechanism link with -ldl. Rename THREADLIBS | |
2156 | to THREADDLLIBS to reflect this. (Thanks to Sven Verdoolaege.) | |
2157 | - Added Hannes Mehnert's patch for FreeBSD/SPARC support. | |
2158 | - Merged some FreeBSD specific patches to threadlibs.c and dyn_load.c. | |
2159 | (Thanks tp John Merryweather Cooper.) | |
2160 | - Define MPROTECT_VDB on MACOSX only if threads are being used, since the | |
2161 | dirty page tracking mechanism uses threads. (This avoids an undefined | |
2162 | reference to _GC_darwin_register_mach_handler_thread.) | |
2163 | - By popular demand, use __libc symbols only if we are built with | |
2164 | USE_LIBC_PRIVATES, which is off by default, and not otherwise documented. | |
2165 | - Ignore GC_enable_incremental() requests when KEEP_BACK_PTRS is set. | |
2166 | The GC itself will dirty lots of pages in this cases, probably making | |
2167 | it counterproductive on all platforms. And the DARWIN port crashes. | |
2168 | ||
2169 | Since gc6.4: | |
2170 | - Integrated Paolo Molaro's patch to deal with EINTR in sem_wait. | |
2171 | - Make GC_approx_sp() write to dummy location to ensure that stack | |
2172 | is grown here, when sp looks reasonable, rather than later, when | |
2173 | it might look like a bad memory reference. (Problem was never | |
2174 | observed that I know of. But on rereading the code it seemed | |
2175 | dubious.) | |
2176 | - Separate out GC_with_callee_saves_pushed and sometimes call | |
2177 | it from GC_suspend_handler in pthread_stop_world.c. Callee-save | |
2178 | register values sometimes failed to get traced under HP/UX on | |
2179 | PA-RISC. Linux/IA64 had the same problem, though non-stacked | |
2180 | callee-save registers seem to be so rarely used there that nobody | |
2181 | ever noticed. | |
2182 | - Integrated an ancient Darwin powerpc_darwin_machine_dep.s patch | |
2183 | from Andreas Tobler, which I had lost. | |
2184 | - Fix compare_and_exchange implementation for gcc/IA64 to deal with | |
2185 | pickier compiler versions. | |
2186 | - Fixed Itanium 32-bit ABI support (HP/UX). In particular, the | |
2187 | compare_and_exchange implementation didn't consider that possibility. | |
2188 | - Undefine GC_pthread_detach in win32_threads.c. (Thanks to | |
2189 | Tagliapietra Tommaso.) | |
2190 | - Fixed inclusion of frame.h for NETBSD in os_dep.c. | |
2191 | - Applied Dan Bonachea's patch to use mmap on AIX. | |
2192 | - Several fixes to resurrect the Irix port on recent OS versions. | |
2193 | - Change ALPHA to use LINUX_STACKBOTTOM. | |
2194 | - Change SPARC64/LINUX to also use LINUX_STACKBOTTOM. Deal with potential | |
2195 | bad values of __libc_stack_end on that platform. (Thanks to David Miller.) | |
2196 | - Relax gctest to allow larger heap if ALIGN_DOUBLE isn't set. | |
2197 | (Unnecessary in 7.0) | |
2198 | - Force a define of __STDC__=0 for the IBM compiler on AIX, so that | |
2199 | we get prototypes. (Unnecessary in 7.0) | |
2200 | - GC_INIT definition for AIX and CYGWIN referred to DATASTART and DATAEND | |
2201 | which are only defined in private include files. | |
2202 | - Integrated some small gcconfig.h patches from Dan Bonachea. Also | |
2203 | relaxed assertion about FreeBSD stack size in pthread_support.c. | |
2204 | - Integrated Andrew Begel's darwin_stop_world.c patch for 64-bit | |
2205 | support. This may need additional work. | |
2206 | - Avoided potentially infinite recursion in GC_save_callers if | |
2207 | the system backtrace calls malloc. The workaround currently requires | |
2208 | __thread support if this code is used with threads. | |
2209 | - Avoided another similar infinite recursion by conditionally | |
2210 | invoking GC_save_callers in alloc.c. (Thanks to Matthias Andree | |
2211 | for helping to track down both of these.) | |
2212 | - Removed all traces of aix_irix_threads.c. AIX and Irix now use | |
2213 | pthread_support.c and pthread_stop_world.c. The old code appeared | |
2214 | to be unreliable for AIX, and was not regularly maintained. | |
2215 | - On Irix, ignore segments with MA_FETCHOP or MA_NOTCACHED attributed; | |
2216 | they're not always safe to read. | |
2217 | - Fixed a previously vacuous assertion (diagnosed by the SGI compiler) | |
2218 | in GC_remove_from_fl. | |
2219 | - Fix stack_size assertion in GC_pthread_create. | |
2220 | - Fix assertion in GC_steal_mark_stack. | |
2221 | ||
2222 | Since gc6.5: | |
2223 | - Fix CPU count detection for Irix and FreeBSD. (Thanks to Dan Bonachea.) | |
2224 | - Integrate Dan Bonachea's patch for the IBM XLC compiler on Darwin. | |
2225 | - Integrated Andreas Tobler's FreeBSD/PowerPC patch. | |
2226 | - Don't access the GC thread structure from the restart handler. It's | |
2227 | unsafe, since the handler may run too late. (Thanks to Ben Maurer for | |
2228 | tracking this down.) | |
2229 | - Applied Christian Thalinger's patch to change comment syntax in | |
2230 | alpha_mach_dep.S. | |
2231 | - Added test for GC_no_dls in GC_dyld_image_add for DARWIN. (Thanks to | |
2232 | Juan Jose Garcia Ripoli). | |
2233 | - Use LINUX_STACKBOTTOM for Linux/SH and LINUX/ARM. (Thanks to Sugioka | |
2234 | Toshinobu and Christian Thalinger.) | |
2235 | - Rewrote GC_parse_map_entry. This assumed a fixed column layout of | |
2236 | /proc/self/maps on Linux. This ceased to be true about 2 years ago. | |
2237 | The old code is probably quite problemetic with -DREDIRECT_MALLOC. It | |
2238 | is also used by default for IA64, though I haven't seen actual failures | |
2239 | there. | |
2240 | - More consistently define HBLKSIZE to 4096 on 64 bit architectures with | |
2241 | 4K pages. (Thanks to Andrew Haley.) | |
2242 | - With win32 threads, GC_stop_world needs to acquire GC_write_cs. (Thanks | |
2243 | to Ben Hutchings for the observation and patch.) | |
2244 | - Move up struct callinfo declaration to make gcc 4.0.2. happy. | |
2245 | ||
2246 | Since 6.6: | |
2247 | - Add "int" to Solaris "end" and "etext" declaration in gc.h. Declared | |
2248 | the symbols with underscores and as arrays, since that's what's actually | |
2249 | used. Perhaps this could all just be removed? (Thanks to John Bowman.) | |
2250 | - Fixed ARM GC_test_and_set code. (Thanks to Kazu Hirata and Paul Brook.) | |
2251 | - Added casts for assignments to hb_last_reclaimed, which truncate the | |
2252 | value. Added a cast to GC_adj_words_allocd. Use GetModuleHandleA | |
2253 | when retrieving a handle to kernel32.dll under win32. (Thanks to the | |
2254 | Visual Prolog developers.) | |
2255 | - Added Tandem S-Series support. (Thanks to Craig McDaniel. A modified | |
2256 | version of his patch was applied, and hence breakage is probably not | |
2257 | his fault.) | |
2258 | - Remove spurious gc:: qualifier for operator delete[] in gc_cpp.h. | |
2259 | (Thanks to Hanno Boeck.) | |
2260 | - Changed a test for LINUX in config_macros.h to one for __linux__. | |
2261 | - Add prototypes for GC_finalizer_notifier and GC_thr_init. (Thanks to | |
2262 | David Ayers.) | |
2263 | - Use ld instead of nonexitent ldz instruction in Darwin FindTopOfStack. | |
2264 | (Thanks to Andreas Tobler.) | |
2265 | - Add support for Darwin/X86. (Thanks to Geoff Norton and the Mono | |
2266 | developers.) | |
2267 | - Merge in some recent gcc fixes. Add ppc64 asm code. (Thanks to Bryce | |
2268 | McKinley and other gcj developers.) | |
2269 | - Scan MEM_PRIVATE sections under Windows ME and predecessors. | |
2270 | - Interior pointers with some largish offsets into large objects could | |
2271 | be ignored, if GC_all_interior_pointers was set. (Oddly this worked | |
2272 | correctly for stack references if it was not set. Otherwise it failed | |
2273 | for both stack and heap references.) Thanks to Andrew McKinlay for the | |
2274 | critical test case. | |
2275 | - Integrated Tatsuya Bizenn's NETBSD threads support, with some | |
2276 | untested changes. | |
2277 | - Added GC_strdup and friends to make leak detection work correctly | |
2278 | for strdup clients. (Thanks to Jon Moore.) Fixed the existing strdup | |
2279 | with malloc redirection to handle a null malloc return correctly. | |
2280 | ||
2281 | Since gc6.7: | |
2282 | - Added some support for Dragonfly BSD. (Thanks to Joerg Sonnenberger and | |
2283 | Thomas Klausner.) | |
2284 | - Improvements to the HP/UX section of configure.in/configure.ac. | |
2285 | (Thanks to Andreas Tobler) | |
2286 | - GC_unix_get_mem could neglect to release the malloc lock on Irix, under | |
2287 | extremely unlikely circumstances. Thanks to Jean-Baptiste Nivois for | |
2288 | some careful code reading. | |
2289 | - Added support for kFreeBSD + glibc (Thanks to Petr Salinger) | |
2290 | - Fix more MacOS threads memory leaks (Thanks to Allan Hsu) | |
2291 | - Added initial Solaris/X86-64 support (Thanks to Rainer Orth) | |
2292 | ||
2293 | Since gc6.8: | |
2294 | - Fix typo in PREFETCH implementation for X86_64. (Thanks to Peter Wang.) | |
2295 | - Fix M68K LINUX port. (Thanks to Debian packagers.) | |
2296 | - __GNUC__ was misspelled as __GNUC in new_gc_alloc.h. (Thanks to Peter Wang.) | |
2297 | - Integrated Allan Hsu's patch for OS X VM deallocation problems. | |
2298 | - Applied FreeBSD/X86_64 patch. | |
2299 | ||
2300 | Since gc6.9: | |
2301 | - Remove GC_PROTO, VOLATILE, GC_PTR, and GC_CONST. Assume ANSI C compiler | |
2302 | and use ANSI constructs unconditionally. | |
2303 | - Introduce #elif and #error in some of the appropriate places. | |
2304 | - Remove GC_printf cruft. Use stdargs. | |
2305 | - Remove separate Solaris threads support. Use the more generic Posix | |
2306 | implementation. | |
2307 | - Use atomic_ops for atomic operations and memory barriers. | |
2308 | - Clean up MPROTECT_VDB implementation. Use SA_SIGINFO wherever | |
2309 | possible. | |
2310 | - Remove broken SIGNALS stuff. | |
2311 | - Use size_t instead of word, where appropriate. | |
2312 | - Add .S.o rule to Makefile.am. | |
2313 | - Officially unsupport SunOS4, several old flavors of M68K (SunOS4, | |
2314 | A/UX, HP), IBM PC/RTs and RISCOS/Irix4. (I doubt the old code worked. | |
2315 | If anyone cares, these should be easy to resurrect.) | |
2316 | - Add EXPECT() in some critical places. | |
2317 | - Redefined hb_sz and hb_body to deal with bytes rather than words. | |
2318 | This affected a great deal of code. I would like to consistently use | |
2319 | byte offsets and sizes where there's not a convincing reason to do | |
2320 | otherwise. | |
2321 | - Redefined several other variables (GC_mem_found, GC_words_allocd) | |
2322 | etc. to use units of bytes. Most of these were also renamed to | |
2323 | reflect that fact. | |
2324 | - Killed as many "register" declarations as possible. | |
2325 | - Partially replaced stubborn allocation with manual write barrier. | |
2326 | It's currently broken. | |
2327 | - Restructured mark code, to allow mark bits to be kept either on | |
2328 | a per allocation granule or per object basis. The emphasis is | |
2329 | now on the -DUSE_MARK_BYTES option, since individual bits perform | |
2330 | quite badly on hyperthreaded P4s, and are probably suboptimal on | |
2331 | other architectures. -DUSE_MARK_BITS is currently broken, and may | |
2332 | be resurrected only for the single-threaded case. This significantly | |
2333 | reduced the cache footprint required by auxiliary GC data structures. | |
2334 | It also reduces space overhead for small heaps. It probably slows | |
2335 | things down slightly if interior pointers are very common. | |
2336 | - As part of the above, we now maintain an approximate count of set | |
2337 | mark bits in each heap block. | |
2338 | - As part of the above, the semantics of hb_map changed drastically. | |
2339 | For MARK_BIT_PER_OBJ, it doesn't exist. For MARK_BIT_PER_GRANULE, | |
2340 | it is purely a way to replace a mod instruction with a table lookup. | |
2341 | (Somewhat to my surprise, this still wins on modern hardware.) | |
2342 | - Removed PRINTSTATS, GATHERSTATS, and SILENT macros. Everything is | |
2343 | now controlled by GC_print_stats variable and GC_PRINT_STATS | |
2344 | and new GC_PRINT_VERBOSE_STATS environment variables. | |
2345 | - Add GC_log_printf and use it consistently for logging output. | |
2346 | - Unconditionally count the objects we reclaim in the sweep phase. | |
2347 | For thread local allocation, we need that anyway, and we expect | |
2348 | that's increasingly the only case that matters. And it simplifies | |
2349 | the code. In general expect minor performance hacks that benefit | |
2350 | only the single-threaded case to disappear. | |
2351 | - Remove GC_quiet from gc.h and elsewhere. | |
2352 | - Changed the heap expansion heuristic, and the definition of | |
2353 | GC_free_space_divisor, to refer to live data size, instead of total | |
2354 | heap size. I believe this is much more robust. It wasn't previously | |
2355 | possible, because we didn't have access to live data size. | |
2356 | - Thread local allocation added the extra byte in twice: Once in | |
2357 | thread_local_alloc, and once in malloc_many. | |
2358 | - Removed GC_malloc_words_small and GC_gcj_fast_malloc. A new | |
2359 | mechanism based on the thread local allocation data structures | |
2360 | is expected to be added instead. This should allow inlined code | |
2361 | that is both fast and doesn't rely on collector internals. | |
2362 | - Changed both free lists and reclaim lists to be indexed by granules | |
2363 | instead of words, norming halving their size. | |
2364 | - MERGE_SIZE is now the only option, and the macro was removed. | |
2365 | (Without it, we need a memory reference to GC_all_interior_pointers | |
2366 | anyway. Thus it costs us nothing.) | |
2367 | - Change GC_size_map to map to granules instead of words. Make sure | |
2368 | that every possible size up to TINY_FREELISTS is present. | |
2369 | - Split of macros need for fast inline allocation into gc_tiny_fl.h | |
2370 | in anticipation of a new inline allocator that doesn't rely on GC | |
2371 | internals. | |
2372 | - Changed thread local allocation to use GRANULE_BYTES and TINY_FREELISTS | |
2373 | in anticipation of a merge with the inline allocation code. | |
2374 | - Removed ALIGN_DOUBLE. This is mostly handled by GRANULE_BYTES. | |
2375 | - Make locking on most platforms conditional on GC_need_to_lock. | |
2376 | ||
2377 | Since gc7.0alpha1: | |
2378 | - GC_bytes_allocd was incremented by a possibly uninitialized variable | |
2379 | in GC_generic_malloc_inner. (Bug introduced in gc7.0alpha1. Thanks | |
2380 | to Ben Hutchings for tracking it down.) | |
2381 | - Win32 fixes. (Thanks to Ben Hutchings and Maurizio Vairani.) | |
2382 | - Integrated Ben Hutchings' GetWriteWatch-based virtual dirty bit | |
2383 | implementation for win32. | |
2384 | - Removed pc_gc.tar and floppy targets in Makefile.direct. Removed | |
2385 | pc_excludes file. | |
2386 | - No longer include GC_bytes_wasted when evaluating allocation progress. | |
2387 | Since we are now counting live memory, it no longer makes sense. | |
2388 | - Applied Davide Angelocola's configury patch. There are now separate | |
2389 | Makefile.am's in the cord and tests subdirectory, more tests, etc. | |
2390 | - Renamed configure.in to configure.ac. | |
2391 | - Merged a very small number of Nathanael Nerode's configure.ac | |
2392 | cleanups from the gcc tree. Unfortunately, that file is a bit | |
2393 | different from ours. | |
2394 | - Changed EINTR handling in sem_wait slightly. | |
2395 | - Restructure the root marking code. Remove all traces of | |
2396 | USE_GENERIC_PUSH_REGS, and effectively make it the default. | |
2397 | Make it easier to pass a context pointer to the mark routine, in | |
2398 | case we ever want to do precise stack marking. | |
2399 | - Replace GC_start_blocking() and GC_end_blocking() with GC_do_blocking(). | |
2400 | This remains undocumented, and only implemented for pthreads. But it | |
2401 | removes an otherwise unavoidable race with stores of callee-save | |
2402 | registers. | |
2403 | - Fix GC_n_mark_bits for the default MARK_BIT_PER_GRANULE case. This | |
2404 | resulted in bogus complaints in heap dumps. | |
2405 | - Upgrade to libatomic_ops-1.0, and update build structure to match. | |
2406 | - Remove SRC_M3 support. Clean up lock initialization code in misc.c. | |
2407 | - Removed gc_local_alloc.h. If THREAD_LOCAL_ALLOC is defined, the | |
2408 | thread local allocation routines are now called automatically. | |
2409 | - Renamed gc_inl.h back to gc_inline.h. Changed the interface appreciably | |
2410 | since locking has turned into a dominant issue, and in-line allocation | |
2411 | only makes sense if it's no worse than thread-local allocation. | |
2412 | Gc_inline.h is now also used to implement thread-local allocation. | |
2413 | - Finished replacing stubborn allocation with manual write barrier. | |
2414 | Untested. | |
2415 | - Use thread-local allocation code by default. | |
2416 | - Added GC_register_my_thread and friends for Posix and win32. | |
2417 | - Patch for GWW_VDB from Ben Hutchings. | |
2418 | - Removed explicit THREAD_LOCAL_ALLOC tests, since that now always | |
2419 | redefines GC_malloc. | |
2420 | - Removed now unused AIX memory allocation code. | |
2421 | - Various minor fixes for bugs introduced in 7.0alpha1. | |
2422 | ||
2423 | Since gc7.0alpha2 | |
2424 | - Added support for dlopen-based interception of pthread functions. | |
2425 | This is only half done. The gc.h redefinitions currently interfere. | |
2426 | - Integrated major automake overhaul from Petter Urkedal. | |
2427 | ||
2428 | Since gc7.0alpha3 | |
2429 | (various 6.5, 6.6 changes) | |
2430 | - Removed GC_brief_async_signal_safe_sleep and used atomic_ops instead. | |
2431 | (Thanks to Ben Maurer.) | |
2432 | - Integrated build patches from David Angelocola and Petter Urkedal. | |
2433 | - Fix dynamic-linker-based pthread call redirection. | |
2434 | - Renamed RS6000 to POWERPC/AIX. | |
2435 | - Allow recovery from SIGSEGV in marker on Linux. This works around | |
2436 | a race in thread stack marking if /proc is used to find roots. We do | |
2437 | that by default with malloc redirection and threads. This involved | |
2438 | moving some GC_find_limit and SETJMP related declarations to gc_priv.h. | |
2439 | - Added doc/porting.html file. | |
2440 | - Added ADD_HEAP_GUARD_PAGES for sbrk/*nix platforms to debug extreme | |
2441 | memory overwrite errors. | |
2442 | - Added trivial NO_INCREMENTAL flag to facilitate debugging. | |
2443 | - Added GC_getattr_np-based GC_get_stack_base (untested). | |
2444 | - Separated thread local allocation into a separate file and added the | |
2445 | beginning of win32 support for that. | |
2446 | ||
2447 | Since gc7.0alpha4 | |
2448 | (more 6.6, 6.7 changes) | |
2449 | - Some Solaris fixes, including some more general changes in how | |
2450 | the assembly pieces of mach_dep.c are handled. | |
2451 | - Removed a lot of SOLARIS_THREADS-specific code that was only | |
2452 | needed with the old implementation. This included many (mostly no-op) | |
2453 | versions of GC_is_fresh. | |
2454 | - Don't use atomic_ops in gc_locks.h unless we need threads. | |
2455 | - Fixed USE_MARK_BITS, which is once againthe default without PARALLEL_MARK. | |
2456 | - Removed Solaris GC_INIT hack. It's a workaround for a long dead bug, | |
2457 | and it seemed to be wrong anyway. | |
2458 | - Changed win32_threads.c to require preprocessor-based interception | |
2459 | of thread routines by default. A client call to GC_use_DllMain is | |
2460 | now required to get the old behavior in which DllMain is used to implicitly | |
2461 | register threads. This was doen for uniformity with other platforms, and | |
2462 | because the DllMain solution seemed to require very tricky code which, | |
2463 | at least in the past, imposed hard bounds onthe number of threads. | |
2464 | - Many small changes to make thread support work again on Cygwin. | |
2465 | - Moved definition of allocator lock etc. to pthread_support.c and | |
2466 | win32_threads.c for those two cases. | |
2467 | - Got rid of the FASTLOCK() machinery. It doesn't seem useful on modern | |
2468 | platforms. | |
2469 | - Cleaned up the uncollectable allocation routines, speeding up the | |
2470 | slower paths. The code did enough unnecessary work off the critical path | |
2471 | that the underlying logic was getting hard to extract. | |
2472 | - No longer turn off THREAD_LOCAL_ALLOC with DBG_HDRS_ALL. Indications | |
2473 | are it just works, and I think the reasons for it not working disappeared | |
2474 | a while ago. | |
2475 | - Fixed bugs in hb_n_marks calculation and assertion. | |
2476 | - Don't use __builtin_expect for pre-3.0 gcc. | |
2477 | - Define GWW_VDB only for recent Microsoft tool chains. | |
2478 | - Add overview.html to doc directory. | |
2479 | - Fix NT_STATIC_THREADS_MAKEFILE, various compiler warnings. | |
2480 | - Made thread local allocation sort of work with Cygwin. The code should | |
2481 | be there to deal with other Windows variants, But non-Cygwin Windows | |
2482 | threads need more bug fixes. | |
2483 | ||
2484 | Since gc7.0alpha5 | |
2485 | (more 6.7 changes) | |
2486 | - Declare GC_dump() in gc.h. | |
2487 | - Add --enable-large-config, which just defines the LARGE_CONFIG macro. | |
2488 | - Make GlobalAlloc address alignment a bit more intuitive. (Thanks to | |
2489 | Charles Mills.) | |
2490 | - Use #elif in the definitions of GET_MEM. | |
2491 | - Overhaul porting.html. Remove corresponding text from README. | |
2492 | - Fix typo in DARWIN section of gcconfig.h. | |
2493 | - Fix Darwin thread memory leak. (Thanks to Bruce Mitchener.) | |
2494 | - Update x86 AO_test_and_set implementation to use "=q". | |
2495 | - Add $(EXEEXT) to many tests in tests/tests.am. (Corresponds to a | |
2496 | 6.7 fix, which no longer applied.) | |
2497 | - Fix Darwin/PPC port. | |
2498 | - Fix Cygwin/threads port. | |
2499 | - Fix gcj malloc support. | |
2500 | - For GNU-style make, don't build libatomic_ops unless threads are requested. | |
2501 | This should allow single-threaded builds on platforms which do not | |
2502 | currently support libatomic_ops. | |
2503 | - Clean up and hopefully fix the CFLAGS calculation for GNU build. | |
2504 | (Substantially improves things on HP/UX.) | |
2505 | - Integrated Andrei Polushin's Visual C++ patches. These provide for | |
2506 | stack traces, better C++ debug support, and better log file handling. | |
2507 | Note that these change the location of the log file to a the path of the | |
2508 | executable with a .log extension. To get the old behavior back, define | |
2509 | OLD_WIN32_LOG_FILE. For the time being, I'm checking his project | |
2510 | files and the like into a windows-untested subdirectory. They | |
2511 | are almost certainly already out of date, but better than what we had | |
2512 | before. | |
2513 | - Fixed some win32 threads bugs, and added support for _beginthreadex. | |
2514 | - Fix zero size thread local allocation so that explicit deallocation | |
2515 | works correctly. | |
2516 | - Removed serious bug in GC_malloc_uncollectable(large size). | |
2517 | - Do not try to do thread-local gcj allocation in incremental mode. There | |
2518 | are races in setting up the descriptor. | |
2519 | - Add GC_INIT() to middle.c, fix some more GC_printfn calls. | |
2520 | - Some assertions erroneously used I_HOLD_LOCK() negatively, eventhough | |
2521 | it can now spuriously return TRUE. | |
2522 | - Rename SUNOS5 macro and OS name to SOLARIS and SUNOS5DL to SOLARISDL. | |
2523 | - On Linux and some Un*x variants, allocate memory by first trying sbrk, | |
2524 | and then switching to mmap if that fails. | |
2525 | - Fixed /proc/x/maps reading to deal with asynchronous deletions. | |
2526 | - Fix REDIRECT_MALLOC with threads on Linux. It now usually seems to work | |
2527 | with ugly hacks that include having calloc behave differently when it is | |
2528 | called from ld.so or the pthreads library. A reasonable amount of | |
2529 | infrastructure was added to support some of this. Thanks to Roland McGrath | |
2530 | for ideas and information. | |
2531 | - Import various updated build scripts. | |
2532 | - Add GC_register_has_static_roots_callback. (Thanks to Andrew Haley.) | |
2533 | - Fix serious bugs in GC_malloc_atomic_uncollectable(). | |
2534 | - Return GC_SUCCESS form GC_get_stack_base(). | |
2535 | - Fix several atomic_ops problems on IA64 with HP Compiler. | |
2536 | - Update to atomic_ops-1.2. | |
2537 | - Fix hb_n_marks description and reclaim.c assertion. | |
2538 | - Various additional win32 threads fixes. | |
2539 | - Enable GC_ASSERTIONS for Debug build with NT_THREADS_MAKEFILE. | |
2540 | ||
2541 | [gc7.0alpha7 was released and version bumped to gc7.0alpha8] | |
2542 | ||
2543 | Since first gc7.0alpha8 version: | |
2544 | [ Some gc6.9 changes ] | |
2545 | - Change FindTopOfStack decl in darwin_stop_world.c. | |
2546 | - Move some static tests from misc.c to gcconfig.h. Use #error. | |
2547 | - Add GC_print_free_list() function. (Thanks to Bruce Hoult.) | |
2548 | - Add GC_GNU_THREADS support on HURD. (Thanks to Aleksey Demakov, | |
2549 | Barry DeFreese, and possibly other Debian maintainers.) | |
2550 | - __GNUC__ was misspelled as __GNUC in thread_local_alloc.h. | |
2551 | (Thanks to Peter Wang.) | |
2552 | - Integrated various MacOSX patches and tried to reconcile them. | |
2553 | Thanks to Allan Hsu, several contributers at Apple, and probably | |
2554 | others. | |
2555 | - Added some casts to powerpc.h in libatomic_ops to silence warnings. | |
2556 | ||
2557 | FOR FURTHER UPDATES SEE ../ChangeLog FILE. | |
2558 | ||
2559 | To do: | |
2560 | - REDIRECT_MALLOC and threads combination should work on more platforms, | |
2561 | and needs more testing on Linux. | |
2562 | - Clone marker inner loop to support arch-dependent prefetching, | |
2563 | and counting of objects marked for finalization. | |
2564 | - The USE_MUNMAP code should really use a separate data structure | |
2565 | indexed by physical page to keep track of time since last use of | |
2566 | a page. Using hblk headers means we lose track of ages when | |
2567 | blocks are merged, and we can't unmap pages that have been allocated and | |
2568 | dropped by the blacklisting code. I suspect both of these matter. | |
2569 | - A dynamic libgc.so references dlopen unconditionally, but doesn't link | |
2570 | against libdl. | |
2571 | - GC_proc_fd for Solaris is not correctly updated in response to a | |
2572 | fork() call. Thus incremental collection in the child won't work | |
2573 | correctly. (Thanks to Ben Cottrell for pointing this out.) | |
2574 | - --enable-redirect-malloc is mostly untested and known not to work | |
2575 | on some platforms. | |
2576 | - There seem to be outstanding issues on Solaris/X86, possibly with | |
2577 | finding the data segment starting address. Information/patches would | |
2578 | be appreciated. | |
2579 | - Very large root set sizes (> 16 MB or so) could cause the collector | |
2580 | to abort with an unexpected mark stack overflow. (Thanks again to | |
2581 | Peter Chubb.) NOT YET FIXED. Workaround is to increase the initial | |
2582 | size. | |
2583 | - The SGI version of the collector marks from mmapped pages, even | |
2584 | if they are not part of dynamic library static data areas. This | |
2585 | causes performance problems with some SGI libraries that use mmap | |
2586 | as a bitmap allocator. NOT YET FIXED. It may be possible to turn | |
2587 | off DYNAMIC_LOADING in the collector as a workaround. It may also | |
2588 | be possible to conditionally intercept mmap and use GC_exclude_static_roots. | |
2589 | The real fix is to walk rld data structures, which looks possible. | |
2590 | - Incremental collector should handle large objects better. Currently, | |
2591 | it looks like the whole object is treated as dirty if any part of it | |
2592 | is. | |
2593 | ||
2594 | FOR FURTHER UPDATES SEE ../ChangeLog FILE. |