src/tools/tidy/src/pal.rs

   1 //! Tidy check to enforce rules about platform-specific code in std.
   2 //!
   3 //! This is intended to maintain existing standards of code
   4 //! organization in hopes that the standard library will continue to
   5 //! be refactored to isolate platform-specific bits, making porting
   6 //! easier; where "standard library" roughly means "all the
   7 //! dependencies of the std and test crates".
   8 //!
   9 //! This generally means placing restrictions on where `cfg(unix)`,
  10 //! `cfg(windows)`, `cfg(target_os)` and `cfg(target_env)` may appear,
  11 //! the basic objective being to isolate platform-specific code to the
  12 //! platform-specific `std::sys` modules, and to the allocation,
  13 //! unwinding, and libc crates.
  14 //!
  15 //! Following are the basic rules, though there are currently
  16 //! exceptions:
  17 //!
  18 //! - core may not have platform-specific code.
  19 //! - libpanic_abort may have platform-specific code.
  20 //! - libpanic_unwind may have platform-specific code.
  21 //! - libunwind may have platform-specific code.
  22 //! - other crates in the std facade may not.
  23 //! - std may have platform-specific code in the following places:
  24 //!   - `sys/unix/`
  25 //!   - `sys/windows/`
  26 //!   - `os/`
  27 //!
  28 //! `std/sys_common` should _not_ contain platform-specific code.
  29 //! Finally, because std contains tests with platform-specific
  30 //! `ignore` attributes, once the parser encounters `mod tests`,
  31 //! platform-specific cfgs are allowed. Not sure yet how to deal with
  32 //! this in the long term.
  33
  34 use std::iter::Iterator;
  35 use std::path::Path;
  36
  37 // Paths that may contain platform-specific code.
  38 const EXCEPTION_PATHS: &[&str] = &[
  39     // std crates
  40     "src/libpanic_abort",
  41     "src/libpanic_unwind",
  42     "src/libunwind",
  43     // black_box implementation is LLVM-version specific and it uses
  44     // target_os to tell targets with different LLVM-versions apart
  45     // (e.g. `wasm32-unknown-emscripten` vs `wasm32-unknown-unknown`):
  46     "src/libcore/hint.rs",
  47     "src/libstd/sys/", // Platform-specific code for std lives here.
  48     // This has the trailing slash so that sys_common is not excepted.
  49     "src/libstd/os", // Platform-specific public interfaces
  50     "src/rtstartup", // Not sure what to do about this. magic stuff for mingw
  51     // temporary exceptions
  52     "src/libstd/lib.rs",
  53     "src/libstd/path.rs",
  54     "src/libstd/f32.rs",
  55     "src/libstd/f64.rs",
  56     // Integration test for platform-specific run-time feature detection:
  57     "src/libstd/tests/run-time-detect.rs",
  58     "src/libstd/net/test.rs",
  59     "src/libstd/sys_common/mod.rs",
  60     "src/libstd/sys_common/net.rs",
  61     "src/libstd/sys_common/backtrace.rs",
  62     "src/libterm", // Not sure how to make this crate portable, but test crate needs it.
  63     "src/libtest", // Probably should defer to unstable `std::sys` APIs.
  64     "src/libstd/sync/mpsc", // some tests are only run on non-emscripten
  65     // std testing crates, okay for now at least
  66     "src/libcore/tests",
  67     "src/liballoc/tests/lib.rs",
  68     // The `VaList` implementation must have platform specific code.
  69     // The Windows implementation of a `va_list` is always a character
  70     // pointer regardless of the target architecture. As a result,
  71     // we must use `#[cfg(windows)]` to conditionally compile the
  72     // correct `VaList` structure for windows.
  73     "src/libcore/ffi.rs",
  74     // non-std crates
  75     "src/test",
  76     "src/tools",
  77     "src/librustc",
  78     "src/librustdoc",
  79     "src/librustc_ast",
  80     "src/bootstrap",
  81 ];
  82
  83 pub fn check(path: &Path, bad: &mut bool) {
  84     // Sanity check that the complex parsing here works.
  85     let mut saw_target_arch = false;
  86     let mut saw_cfg_bang = false;
  87     super::walk(path, &mut super::filter_dirs, &mut |entry, contents| {
  88         let file = entry.path();
  89         let filestr = file.to_string_lossy().replace("\\", "/");
  90         if !filestr.ends_with(".rs") {
  91             return;
  92         }
  93
  94         let is_exception_path = EXCEPTION_PATHS.iter().any(|s| filestr.contains(&**s));
  95         if is_exception_path {
  96             return;
  97         }
  98
  99         check_cfgs(contents, &file, bad, &mut saw_target_arch, &mut saw_cfg_bang);
 100     });
 101
 102     assert!(saw_target_arch);
 103     assert!(saw_cfg_bang);
 104 }
 105
 106 fn check_cfgs(
 107     contents: &str,
 108     file: &Path,
 109     bad: &mut bool,
 110     saw_target_arch: &mut bool,
 111     saw_cfg_bang: &mut bool,
 112 ) {
 113     // For now it's ok to have platform-specific code after 'mod tests'.
 114     let mod_tests_idx = find_test_mod(contents);
 115     let contents = &contents[..mod_tests_idx];
 116     // Pull out all `cfg(...)` and `cfg!(...)` strings.
 117     let cfgs = parse_cfgs(contents);
 118
 119     let mut line_numbers: Option<Vec<usize>> = None;
 120     let mut err = |idx: usize, cfg: &str| {
 121         if line_numbers.is_none() {
 122             line_numbers = Some(contents.match_indices('\n').map(|(i, _)| i).collect());
 123         }
 124         let line_numbers = line_numbers.as_ref().expect("");
 125         let line = match line_numbers.binary_search(&idx) {
 126             Ok(_) => unreachable!(),
 127             Err(i) => i + 1,
 128         };
 129         tidy_error!(bad, "{}:{}: platform-specific cfg: {}", file.display(), line, cfg);
 130     };
 131
 132     for (idx, cfg) in cfgs {
 133         // Sanity check that the parsing here works.
 134         if !*saw_target_arch && cfg.contains("target_arch") {
 135             *saw_target_arch = true
 136         }
 137         if !*saw_cfg_bang && cfg.contains("cfg!") {
 138             *saw_cfg_bang = true
 139         }
 140
 141         let contains_platform_specific_cfg = cfg.contains("target_os")
 142             || cfg.contains("target_env")
 143             || cfg.contains("target_vendor")
 144             || cfg.contains("unix")
 145             || cfg.contains("windows");
 146
 147         if !contains_platform_specific_cfg {
 148             continue;
 149         }
 150
 151         let preceeded_by_doc_comment = {
 152             let pre_contents = &contents[..idx];
 153             let pre_newline = pre_contents.rfind('\n');
 154             let pre_doc_comment = pre_contents.rfind("///");
 155             match (pre_newline, pre_doc_comment) {
 156                 (Some(n), Some(c)) => n < c,
 157                 (None, Some(_)) => true,
 158                 (_, None) => false,
 159             }
 160         };
 161
 162         if preceeded_by_doc_comment {
 163             continue;
 164         }
 165
 166         err(idx, cfg);
 167     }
 168 }
 169
 170 fn find_test_mod(contents: &str) -> usize {
 171     if let Some(mod_tests_idx) = contents.find("mod tests") {
 172         // Also capture a previous line indicating that "mod tests" is cfg'd out.
 173         let prev_newline_idx = contents[..mod_tests_idx].rfind('\n').unwrap_or(mod_tests_idx);
 174         let prev_newline_idx = contents[..prev_newline_idx].rfind('\n');
 175         if let Some(nl) = prev_newline_idx {
 176             let prev_line = &contents[nl + 1..mod_tests_idx];
 177             if prev_line.contains("cfg(all(test, not(target_os")
 178                 || prev_line.contains("cfg(all(test, not(any(target_os")
 179             {
 180                 nl
 181             } else {
 182                 mod_tests_idx
 183             }
 184         } else {
 185             mod_tests_idx
 186         }
 187     } else {
 188         contents.len()
 189     }
 190 }
 191
 192 fn parse_cfgs<'a>(contents: &'a str) -> Vec<(usize, &'a str)> {
 193     let candidate_cfgs = contents.match_indices("cfg");
 194     let candidate_cfg_idxs = candidate_cfgs.map(|(i, _)| i);
 195     // This is puling out the indexes of all "cfg" strings
 196     // that appear to be tokens followed by a parenthesis.
 197     let cfgs = candidate_cfg_idxs.filter(|i| {
 198         let pre_idx = i.saturating_sub(*i);
 199         let succeeds_non_ident = !contents
 200             .as_bytes()
 201             .get(pre_idx)
 202             .cloned()
 203             .map(char::from)
 204             .map(char::is_alphanumeric)
 205             .unwrap_or(false);
 206         let contents_after = &contents[*i..];
 207         let first_paren = contents_after.find('(');
 208         let paren_idx = first_paren.map(|ip| i + ip);
 209         let preceeds_whitespace_and_paren = paren_idx
 210             .map(|ip| {
 211                 let maybe_space = &contents[*i + "cfg".len()..ip];
 212                 maybe_space.chars().all(|c| char::is_whitespace(c) || c == '!')
 213             })
 214             .unwrap_or(false);
 215
 216         succeeds_non_ident && preceeds_whitespace_and_paren
 217     });
 218
 219     cfgs.flat_map(|i| {
 220         let mut depth = 0;
 221         let contents_from = &contents[i..];
 222         for (j, byte) in contents_from.bytes().enumerate() {
 223             match byte {
 224                 b'(' => {
 225                     depth += 1;
 226                 }
 227                 b')' => {
 228                     depth -= 1;
 229                     if depth == 0 {
 230                         return Some((i, &contents_from[..=j]));
 231                     }
 232                 }
 233                 _ => {}
 234             }
 235         }
 236
 237         // if the parentheses are unbalanced just ignore this cfg -- it'll be caught when attempting
 238         // to run the compiler, and there's no real reason to lint it separately here
 239         None
 240     })
 241     .collect()
 242 }