]> git.proxmox.com Git - rustc.git/blobdiff - vendor/rustc-ap-rustc_session/src/session.rs
New upstream version 1.52.1+dfsg1
[rustc.git] / vendor / rustc-ap-rustc_session / src / session.rs
diff --git a/vendor/rustc-ap-rustc_session/src/session.rs b/vendor/rustc-ap-rustc_session/src/session.rs
new file mode 100644 (file)
index 0000000..69aa72d
--- /dev/null
@@ -0,0 +1,1639 @@
+use crate::cgu_reuse_tracker::CguReuseTracker;
+use crate::code_stats::CodeStats;
+pub use crate::code_stats::{DataTypeKind, FieldInfo, SizeKind, VariantInfo};
+use crate::config::{self, CrateType, OutputType, PrintRequest, SanitizerSet, SwitchWithOptPath};
+use crate::filesearch;
+use crate::lint::{self, LintId};
+use crate::parse::ParseSess;
+use crate::search_paths::{PathKind, SearchPath};
+
+pub use rustc_ast::attr::MarkedAttrs;
+pub use rustc_ast::crate_disambiguator::CrateDisambiguator;
+pub use rustc_ast::Attribute;
+use rustc_data_structures::flock;
+use rustc_data_structures::fx::{FxHashMap, FxHashSet};
+use rustc_data_structures::jobserver::{self, Client};
+use rustc_data_structures::profiling::{duration_to_secs_str, SelfProfiler, SelfProfilerRef};
+use rustc_data_structures::sync::{
+    self, AtomicU64, AtomicUsize, Lock, Lrc, OnceCell, OneThread, Ordering, Ordering::SeqCst,
+};
+use rustc_errors::annotate_snippet_emitter_writer::AnnotateSnippetEmitterWriter;
+use rustc_errors::emitter::{Emitter, EmitterWriter, HumanReadableErrorType};
+use rustc_errors::json::JsonEmitter;
+use rustc_errors::registry::Registry;
+use rustc_errors::{Applicability, Diagnostic, DiagnosticBuilder, DiagnosticId, ErrorReported};
+use rustc_lint_defs::FutureBreakage;
+use rustc_span::edition::Edition;
+use rustc_span::source_map::{FileLoader, MultiSpan, RealFileLoader, SourceMap, Span};
+use rustc_span::{sym, SourceFileHashAlgorithm, Symbol};
+use rustc_target::asm::InlineAsmArch;
+use rustc_target::spec::{CodeModel, PanicStrategy, RelocModel, RelroLevel};
+use rustc_target::spec::{SplitDebuginfo, Target, TargetTriple, TlsModel};
+
+use std::cell::{self, RefCell};
+use std::env;
+use std::fmt;
+use std::io::Write;
+use std::num::NonZeroU32;
+use std::ops::{Div, Mul};
+use std::path::PathBuf;
+use std::str::FromStr;
+use std::sync::Arc;
+use std::time::Duration;
+
+pub trait SessionLintStore: sync::Send + sync::Sync {
+    fn name_to_lint(&self, lint_name: &str) -> LintId;
+}
+
+pub struct OptimizationFuel {
+    /// If `-zfuel=crate=n` is specified, initially set to `n`, otherwise `0`.
+    remaining: u64,
+    /// We're rejecting all further optimizations.
+    out_of_fuel: bool,
+}
+
+/// The behavior of the CTFE engine when an error occurs with regards to backtraces.
+#[derive(Clone, Copy)]
+pub enum CtfeBacktrace {
+    /// Do nothing special, return the error as usual without a backtrace.
+    Disabled,
+    /// Capture a backtrace at the point the error is created and return it in the error
+    /// (to be printed later if/when the error ever actually gets shown to the user).
+    Capture,
+    /// Capture a backtrace at the point the error is created and immediately print it out.
+    Immediate,
+}
+
+/// New-type wrapper around `usize` for representing limits. Ensures that comparisons against
+/// limits are consistent throughout the compiler.
+#[derive(Clone, Copy, Debug)]
+pub struct Limit(pub usize);
+
+impl Limit {
+    /// Create a new limit from a `usize`.
+    pub fn new(value: usize) -> Self {
+        Limit(value)
+    }
+
+    /// Check that `value` is within the limit. Ensures that the same comparisons are used
+    /// throughout the compiler, as mismatches can cause ICEs, see #72540.
+    pub fn value_within_limit(&self, value: usize) -> bool {
+        value <= self.0
+    }
+}
+
+impl fmt::Display for Limit {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{}", self.0)
+    }
+}
+
+impl Div<usize> for Limit {
+    type Output = Limit;
+
+    fn div(self, rhs: usize) -> Self::Output {
+        Limit::new(self.0 / rhs)
+    }
+}
+
+impl Mul<usize> for Limit {
+    type Output = Limit;
+
+    fn mul(self, rhs: usize) -> Self::Output {
+        Limit::new(self.0 * rhs)
+    }
+}
+
+/// Represents the data associated with a compilation
+/// session for a single crate.
+pub struct Session {
+    pub target: Target,
+    pub host: Target,
+    pub opts: config::Options,
+    pub host_tlib_path: SearchPath,
+    /// `None` if the host and target are the same.
+    pub target_tlib_path: Option<SearchPath>,
+    pub parse_sess: ParseSess,
+    pub sysroot: PathBuf,
+    /// The name of the root source file of the crate, in the local file system.
+    /// `None` means that there is no source file.
+    pub local_crate_source_file: Option<PathBuf>,
+    /// The directory the compiler has been executed in plus a flag indicating
+    /// if the value stored here has been affected by path remapping.
+    pub working_dir: (PathBuf, bool),
+
+    /// Set of `(DiagnosticId, Option<Span>, message)` tuples tracking
+    /// (sub)diagnostics that have been set once, but should not be set again,
+    /// in order to avoid redundantly verbose output (Issue #24690, #44953).
+    pub one_time_diagnostics: Lock<FxHashSet<(DiagnosticMessageId, Option<Span>, String)>>,
+    crate_types: OnceCell<Vec<CrateType>>,
+    /// The `crate_disambiguator` is constructed out of all the `-C metadata`
+    /// arguments passed to the compiler. Its value together with the crate-name
+    /// forms a unique global identifier for the crate. It is used to allow
+    /// multiple crates with the same name to coexist. See the
+    /// `rustc_codegen_llvm::back::symbol_names` module for more information.
+    pub crate_disambiguator: OnceCell<CrateDisambiguator>,
+
+    features: OnceCell<rustc_feature::Features>,
+
+    lint_store: OnceCell<Lrc<dyn SessionLintStore>>,
+
+    /// The maximum recursion limit for potentially infinitely recursive
+    /// operations such as auto-dereference and monomorphization.
+    pub recursion_limit: OnceCell<Limit>,
+
+    /// The maximum length of types during monomorphization.
+    pub type_length_limit: OnceCell<Limit>,
+
+    /// The maximum blocks a const expression can evaluate.
+    pub const_eval_limit: OnceCell<Limit>,
+
+    incr_comp_session: OneThread<RefCell<IncrCompSession>>,
+    /// Used for incremental compilation tests. Will only be populated if
+    /// `-Zquery-dep-graph` is specified.
+    pub cgu_reuse_tracker: CguReuseTracker,
+
+    /// Used by `-Z self-profile`.
+    pub prof: SelfProfilerRef,
+
+    /// Some measurements that are being gathered during compilation.
+    pub perf_stats: PerfStats,
+
+    /// Data about code being compiled, gathered during compilation.
+    pub code_stats: CodeStats,
+
+    /// If `-zfuel=crate=n` is specified, `Some(crate)`.
+    optimization_fuel_crate: Option<String>,
+
+    /// Tracks fuel info if `-zfuel=crate=n` is specified.
+    optimization_fuel: Lock<OptimizationFuel>,
+
+    // The next two are public because the driver needs to read them.
+    /// If `-zprint-fuel=crate`, `Some(crate)`.
+    pub print_fuel_crate: Option<String>,
+    /// Always set to zero and incremented so that we can print fuel expended by a crate.
+    pub print_fuel: AtomicU64,
+
+    /// Loaded up early on in the initialization of this `Session` to avoid
+    /// false positives about a job server in our environment.
+    pub jobserver: Client,
+
+    /// Cap lint level specified by a driver specifically.
+    pub driver_lint_caps: FxHashMap<lint::LintId, lint::Level>,
+
+    /// `Span`s of trait methods that weren't found to avoid emitting object safety errors
+    pub trait_methods_not_found: Lock<FxHashSet<Span>>,
+
+    /// Mapping from ident span to path span for paths that don't exist as written, but that
+    /// exist under `std`. For example, wrote `str::from_utf8` instead of `std::str::from_utf8`.
+    pub confused_type_with_std_module: Lock<FxHashMap<Span, Span>>,
+
+    /// Path for libraries that will take preference over libraries shipped by Rust.
+    /// Used by windows-gnu targets to priortize system mingw-w64 libraries.
+    pub system_library_path: OneThread<RefCell<Option<Option<PathBuf>>>>,
+
+    /// Tracks the current behavior of the CTFE engine when an error occurs.
+    /// Options range from returning the error without a backtrace to returning an error
+    /// and immediately printing the backtrace to stderr.
+    pub ctfe_backtrace: Lock<CtfeBacktrace>,
+
+    /// This tracks where `-Zunleash-the-miri-inside-of-you` was used to get around a
+    /// const check, optionally with the relevant feature gate.  We use this to
+    /// warn about unleashing, but with a single diagnostic instead of dozens that
+    /// drown everything else in noise.
+    miri_unleashed_features: Lock<Vec<(Span, Option<Symbol>)>>,
+
+    /// Base directory containing the `src/` for the Rust standard library, and
+    /// potentially `rustc` as well, if we can can find it. Right now it's always
+    /// `$sysroot/lib/rustlib/src/rust` (i.e. the `rustup` `rust-src` component).
+    ///
+    /// This directory is what the virtual `/rustc/$hash` is translated back to,
+    /// if Rust was built with path remapping to `/rustc/$hash` enabled
+    /// (the `rust.remap-debuginfo` option in `config.toml`).
+    pub real_rust_source_base_dir: Option<PathBuf>,
+
+    /// Architecture to use for interpreting asm!.
+    pub asm_arch: Option<InlineAsmArch>,
+
+    /// Set of enabled features for the current target.
+    pub target_features: FxHashSet<Symbol>,
+
+    known_attrs: Lock<MarkedAttrs>,
+    used_attrs: Lock<MarkedAttrs>,
+
+    /// `Span`s for `if` conditions that we have suggested turning into `if let`.
+    pub if_let_suggestions: Lock<FxHashSet<Span>>,
+}
+
+pub struct PerfStats {
+    /// The accumulated time spent on computing symbol hashes.
+    pub symbol_hash_time: Lock<Duration>,
+    /// Total number of values canonicalized queries constructed.
+    pub queries_canonicalized: AtomicUsize,
+    /// Number of times this query is invoked.
+    pub normalize_generic_arg_after_erasing_regions: AtomicUsize,
+    /// Number of times this query is invoked.
+    pub normalize_projection_ty: AtomicUsize,
+}
+
+/// Enum to support dispatch of one-time diagnostics (in `Session.diag_once`).
+enum DiagnosticBuilderMethod {
+    Note,
+    SpanNote,
+    SpanSuggestion(String), // suggestion
+                            // Add more variants as needed to support one-time diagnostics.
+}
+
+/// Trait implemented by error types. This should not be implemented manually. Instead, use
+/// `#[derive(SessionDiagnostic)]` -- see [rustc_macros::SessionDiagnostic].
+pub trait SessionDiagnostic<'a> {
+    /// Write out as a diagnostic out of `sess`.
+    #[must_use]
+    fn into_diagnostic(self, sess: &'a Session) -> DiagnosticBuilder<'a>;
+}
+
+/// Diagnostic message ID, used by `Session.one_time_diagnostics` to avoid
+/// emitting the same message more than once.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum DiagnosticMessageId {
+    ErrorId(u16), // EXXXX error code as integer
+    LintId(lint::LintId),
+    StabilityId(Option<NonZeroU32>), // issue number
+}
+
+impl From<&'static lint::Lint> for DiagnosticMessageId {
+    fn from(lint: &'static lint::Lint) -> Self {
+        DiagnosticMessageId::LintId(lint::LintId::of(lint))
+    }
+}
+
+impl Session {
+    pub fn miri_unleashed_feature(&self, span: Span, feature_gate: Option<Symbol>) {
+        self.miri_unleashed_features.lock().push((span, feature_gate));
+    }
+
+    fn check_miri_unleashed_features(&self) {
+        let unleashed_features = self.miri_unleashed_features.lock();
+        if !unleashed_features.is_empty() {
+            let mut must_err = false;
+            // Create a diagnostic pointing at where things got unleashed.
+            let mut diag = self.struct_warn("skipping const checks");
+            for &(span, feature_gate) in unleashed_features.iter() {
+                // FIXME: `span_label` doesn't do anything, so we use "help" as a hack.
+                if let Some(feature_gate) = feature_gate {
+                    diag.span_help(span, &format!("skipping check for `{}` feature", feature_gate));
+                    // The unleash flag must *not* be used to just "hack around" feature gates.
+                    must_err = true;
+                } else {
+                    diag.span_help(span, "skipping check that does not even have a feature gate");
+                }
+            }
+            diag.emit();
+            // If we should err, make sure we did.
+            if must_err && !self.has_errors() {
+                // We have skipped a feature gate, and not run into other errors... reject.
+                self.err(
+                    "`-Zunleash-the-miri-inside-of-you` may not be used to circumvent feature \
+                     gates, except when testing error paths in the CTFE engine",
+                );
+            }
+        }
+    }
+
+    /// Invoked all the way at the end to finish off diagnostics printing.
+    pub fn finish_diagnostics(&self, registry: &Registry) {
+        self.check_miri_unleashed_features();
+        self.diagnostic().print_error_count(registry);
+        self.emit_future_breakage();
+    }
+
+    fn emit_future_breakage(&self) {
+        if !self.opts.debugging_opts.emit_future_incompat_report {
+            return;
+        }
+
+        let diags = self.diagnostic().take_future_breakage_diagnostics();
+        if diags.is_empty() {
+            return;
+        }
+        // If any future-breakage lints were registered, this lint store
+        // should be available
+        let lint_store = self.lint_store.get().expect("`lint_store` not initialized!");
+        let diags_and_breakage: Vec<(FutureBreakage, Diagnostic)> = diags
+            .into_iter()
+            .map(|diag| {
+                let lint_name = match &diag.code {
+                    Some(DiagnosticId::Lint { name, has_future_breakage: true }) => name,
+                    _ => panic!("Unexpected code in diagnostic {:?}", diag),
+                };
+                let lint = lint_store.name_to_lint(&lint_name);
+                let future_breakage =
+                    lint.lint.future_incompatible.unwrap().future_breakage.unwrap();
+                (future_breakage, diag)
+            })
+            .collect();
+        self.parse_sess.span_diagnostic.emit_future_breakage_report(diags_and_breakage);
+    }
+
+    pub fn local_crate_disambiguator(&self) -> CrateDisambiguator {
+        self.crate_disambiguator.get().copied().unwrap()
+    }
+
+    pub fn crate_types(&self) -> &[CrateType] {
+        self.crate_types.get().unwrap().as_slice()
+    }
+
+    pub fn init_crate_types(&self, crate_types: Vec<CrateType>) {
+        self.crate_types.set(crate_types).expect("`crate_types` was initialized twice")
+    }
+
+    pub fn recursion_limit(&self) -> Limit {
+        self.recursion_limit.get().copied().unwrap()
+    }
+
+    pub fn type_length_limit(&self) -> Limit {
+        self.type_length_limit.get().copied().unwrap()
+    }
+
+    pub fn const_eval_limit(&self) -> Limit {
+        self.const_eval_limit.get().copied().unwrap()
+    }
+
+    pub fn struct_span_warn<S: Into<MultiSpan>>(&self, sp: S, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_warn(sp, msg)
+    }
+    pub fn struct_span_warn_with_code<S: Into<MultiSpan>>(
+        &self,
+        sp: S,
+        msg: &str,
+        code: DiagnosticId,
+    ) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_warn_with_code(sp, msg, code)
+    }
+    pub fn struct_warn(&self, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_warn(msg)
+    }
+    pub fn struct_span_allow<S: Into<MultiSpan>>(&self, sp: S, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_allow(sp, msg)
+    }
+    pub fn struct_allow(&self, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_allow(msg)
+    }
+    pub fn struct_span_err<S: Into<MultiSpan>>(&self, sp: S, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_err(sp, msg)
+    }
+    pub fn struct_span_err_with_code<S: Into<MultiSpan>>(
+        &self,
+        sp: S,
+        msg: &str,
+        code: DiagnosticId,
+    ) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_err_with_code(sp, msg, code)
+    }
+    // FIXME: This method should be removed (every error should have an associated error code).
+    pub fn struct_err(&self, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_err(msg)
+    }
+    pub fn struct_err_with_code(&self, msg: &str, code: DiagnosticId) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_err_with_code(msg, code)
+    }
+    pub fn struct_span_fatal<S: Into<MultiSpan>>(&self, sp: S, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_fatal(sp, msg)
+    }
+    pub fn struct_span_fatal_with_code<S: Into<MultiSpan>>(
+        &self,
+        sp: S,
+        msg: &str,
+        code: DiagnosticId,
+    ) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_span_fatal_with_code(sp, msg, code)
+    }
+    pub fn struct_fatal(&self, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_fatal(msg)
+    }
+
+    pub fn span_fatal<S: Into<MultiSpan>>(&self, sp: S, msg: &str) -> ! {
+        self.diagnostic().span_fatal(sp, msg).raise()
+    }
+    pub fn span_fatal_with_code<S: Into<MultiSpan>>(
+        &self,
+        sp: S,
+        msg: &str,
+        code: DiagnosticId,
+    ) -> ! {
+        self.diagnostic().span_fatal_with_code(sp, msg, code).raise()
+    }
+    pub fn fatal(&self, msg: &str) -> ! {
+        self.diagnostic().fatal(msg).raise()
+    }
+    pub fn span_err_or_warn<S: Into<MultiSpan>>(&self, is_warning: bool, sp: S, msg: &str) {
+        if is_warning {
+            self.span_warn(sp, msg);
+        } else {
+            self.span_err(sp, msg);
+        }
+    }
+    pub fn span_err<S: Into<MultiSpan>>(&self, sp: S, msg: &str) {
+        self.diagnostic().span_err(sp, msg)
+    }
+    pub fn span_err_with_code<S: Into<MultiSpan>>(&self, sp: S, msg: &str, code: DiagnosticId) {
+        self.diagnostic().span_err_with_code(sp, &msg, code)
+    }
+    pub fn err(&self, msg: &str) {
+        self.diagnostic().err(msg)
+    }
+    pub fn emit_err<'a>(&'a self, err: impl SessionDiagnostic<'a>) {
+        err.into_diagnostic(self).emit()
+    }
+    pub fn err_count(&self) -> usize {
+        self.diagnostic().err_count()
+    }
+    pub fn has_errors(&self) -> bool {
+        self.diagnostic().has_errors()
+    }
+    pub fn has_errors_or_delayed_span_bugs(&self) -> bool {
+        self.diagnostic().has_errors_or_delayed_span_bugs()
+    }
+    pub fn abort_if_errors(&self) {
+        self.diagnostic().abort_if_errors();
+    }
+    pub fn compile_status(&self) -> Result<(), ErrorReported> {
+        if self.has_errors() {
+            self.diagnostic().emit_stashed_diagnostics();
+            Err(ErrorReported)
+        } else {
+            Ok(())
+        }
+    }
+    // FIXME(matthewjasper) Remove this method, it should never be needed.
+    pub fn track_errors<F, T>(&self, f: F) -> Result<T, ErrorReported>
+    where
+        F: FnOnce() -> T,
+    {
+        let old_count = self.err_count();
+        let result = f();
+        let errors = self.err_count() - old_count;
+        if errors == 0 { Ok(result) } else { Err(ErrorReported) }
+    }
+    pub fn span_warn<S: Into<MultiSpan>>(&self, sp: S, msg: &str) {
+        self.diagnostic().span_warn(sp, msg)
+    }
+    pub fn span_warn_with_code<S: Into<MultiSpan>>(&self, sp: S, msg: &str, code: DiagnosticId) {
+        self.diagnostic().span_warn_with_code(sp, msg, code)
+    }
+    pub fn warn(&self, msg: &str) {
+        self.diagnostic().warn(msg)
+    }
+    pub fn opt_span_warn<S: Into<MultiSpan>>(&self, opt_sp: Option<S>, msg: &str) {
+        match opt_sp {
+            Some(sp) => self.span_warn(sp, msg),
+            None => self.warn(msg),
+        }
+    }
+    /// Delay a span_bug() call until abort_if_errors()
+    #[track_caller]
+    pub fn delay_span_bug<S: Into<MultiSpan>>(&self, sp: S, msg: &str) {
+        self.diagnostic().delay_span_bug(sp, msg)
+    }
+
+    /// Used for code paths of expensive computations that should only take place when
+    /// warnings or errors are emitted. If no messages are emitted ("good path"), then
+    /// it's likely a bug.
+    pub fn delay_good_path_bug(&self, msg: &str) {
+        if self.opts.debugging_opts.print_type_sizes
+            || self.opts.debugging_opts.query_dep_graph
+            || self.opts.debugging_opts.dump_mir.is_some()
+            || self.opts.debugging_opts.unpretty.is_some()
+            || self.opts.output_types.contains_key(&OutputType::Mir)
+            || std::env::var_os("RUSTC_LOG").is_some()
+        {
+            return;
+        }
+
+        self.diagnostic().delay_good_path_bug(msg)
+    }
+
+    pub fn note_without_error(&self, msg: &str) {
+        self.diagnostic().note_without_error(msg)
+    }
+    pub fn span_note_without_error<S: Into<MultiSpan>>(&self, sp: S, msg: &str) {
+        self.diagnostic().span_note_without_error(sp, msg)
+    }
+    pub fn struct_note_without_error(&self, msg: &str) -> DiagnosticBuilder<'_> {
+        self.diagnostic().struct_note_without_error(msg)
+    }
+
+    pub fn diagnostic(&self) -> &rustc_errors::Handler {
+        &self.parse_sess.span_diagnostic
+    }
+
+    /// Analogous to calling methods on the given `DiagnosticBuilder`, but
+    /// deduplicates on lint ID, span (if any), and message for this `Session`
+    fn diag_once<'a, 'b>(
+        &'a self,
+        diag_builder: &'b mut DiagnosticBuilder<'a>,
+        method: DiagnosticBuilderMethod,
+        msg_id: DiagnosticMessageId,
+        message: &str,
+        span_maybe: Option<Span>,
+    ) {
+        let id_span_message = (msg_id, span_maybe, message.to_owned());
+        let fresh = self.one_time_diagnostics.borrow_mut().insert(id_span_message);
+        if fresh {
+            match method {
+                DiagnosticBuilderMethod::Note => {
+                    diag_builder.note(message);
+                }
+                DiagnosticBuilderMethod::SpanNote => {
+                    let span = span_maybe.expect("`span_note` needs a span");
+                    diag_builder.span_note(span, message);
+                }
+                DiagnosticBuilderMethod::SpanSuggestion(suggestion) => {
+                    let span = span_maybe.expect("`span_suggestion_*` needs a span");
+                    diag_builder.span_suggestion(
+                        span,
+                        message,
+                        suggestion,
+                        Applicability::Unspecified,
+                    );
+                }
+            }
+        }
+    }
+
+    pub fn diag_span_note_once<'a, 'b>(
+        &'a self,
+        diag_builder: &'b mut DiagnosticBuilder<'a>,
+        msg_id: DiagnosticMessageId,
+        span: Span,
+        message: &str,
+    ) {
+        self.diag_once(
+            diag_builder,
+            DiagnosticBuilderMethod::SpanNote,
+            msg_id,
+            message,
+            Some(span),
+        );
+    }
+
+    pub fn diag_note_once<'a, 'b>(
+        &'a self,
+        diag_builder: &'b mut DiagnosticBuilder<'a>,
+        msg_id: DiagnosticMessageId,
+        message: &str,
+    ) {
+        self.diag_once(diag_builder, DiagnosticBuilderMethod::Note, msg_id, message, None);
+    }
+
+    pub fn diag_span_suggestion_once<'a, 'b>(
+        &'a self,
+        diag_builder: &'b mut DiagnosticBuilder<'a>,
+        msg_id: DiagnosticMessageId,
+        span: Span,
+        message: &str,
+        suggestion: String,
+    ) {
+        self.diag_once(
+            diag_builder,
+            DiagnosticBuilderMethod::SpanSuggestion(suggestion),
+            msg_id,
+            message,
+            Some(span),
+        );
+    }
+
+    #[inline]
+    pub fn source_map(&self) -> &SourceMap {
+        self.parse_sess.source_map()
+    }
+    pub fn verbose(&self) -> bool {
+        self.opts.debugging_opts.verbose
+    }
+    pub fn time_passes(&self) -> bool {
+        self.opts.debugging_opts.time_passes || self.opts.debugging_opts.time
+    }
+    pub fn instrument_mcount(&self) -> bool {
+        self.opts.debugging_opts.instrument_mcount
+    }
+    pub fn time_llvm_passes(&self) -> bool {
+        self.opts.debugging_opts.time_llvm_passes
+    }
+    pub fn meta_stats(&self) -> bool {
+        self.opts.debugging_opts.meta_stats
+    }
+    pub fn asm_comments(&self) -> bool {
+        self.opts.debugging_opts.asm_comments
+    }
+    pub fn verify_llvm_ir(&self) -> bool {
+        self.opts.debugging_opts.verify_llvm_ir || option_env!("RUSTC_VERIFY_LLVM_IR").is_some()
+    }
+    pub fn borrowck_stats(&self) -> bool {
+        self.opts.debugging_opts.borrowck_stats
+    }
+    pub fn print_llvm_passes(&self) -> bool {
+        self.opts.debugging_opts.print_llvm_passes
+    }
+    pub fn binary_dep_depinfo(&self) -> bool {
+        self.opts.debugging_opts.binary_dep_depinfo
+    }
+
+    /// Gets the features enabled for the current compilation session.
+    /// DO NOT USE THIS METHOD if there is a TyCtxt available, as it circumvents
+    /// dependency tracking. Use tcx.features() instead.
+    #[inline]
+    pub fn features_untracked(&self) -> &rustc_feature::Features {
+        self.features.get().unwrap()
+    }
+
+    pub fn init_features(&self, features: rustc_feature::Features) {
+        match self.features.set(features) {
+            Ok(()) => {}
+            Err(_) => panic!("`features` was initialized twice"),
+        }
+    }
+
+    pub fn init_lint_store(&self, lint_store: Lrc<dyn SessionLintStore>) {
+        self.lint_store
+            .set(lint_store)
+            .map_err(|_| ())
+            .expect("`lint_store` was initialized twice");
+    }
+
+    /// Calculates the flavor of LTO to use for this compilation.
+    pub fn lto(&self) -> config::Lto {
+        // If our target has codegen requirements ignore the command line
+        if self.target.requires_lto {
+            return config::Lto::Fat;
+        }
+
+        // If the user specified something, return that. If they only said `-C
+        // lto` and we've for whatever reason forced off ThinLTO via the CLI,
+        // then ensure we can't use a ThinLTO.
+        match self.opts.cg.lto {
+            config::LtoCli::Unspecified => {
+                // The compiler was invoked without the `-Clto` flag. Fall
+                // through to the default handling
+            }
+            config::LtoCli::No => {
+                // The user explicitly opted out of any kind of LTO
+                return config::Lto::No;
+            }
+            config::LtoCli::Yes | config::LtoCli::Fat | config::LtoCli::NoParam => {
+                // All of these mean fat LTO
+                return config::Lto::Fat;
+            }
+            config::LtoCli::Thin => {
+                return if self.opts.cli_forced_thinlto_off {
+                    config::Lto::Fat
+                } else {
+                    config::Lto::Thin
+                };
+            }
+        }
+
+        // Ok at this point the target doesn't require anything and the user
+        // hasn't asked for anything. Our next decision is whether or not
+        // we enable "auto" ThinLTO where we use multiple codegen units and
+        // then do ThinLTO over those codegen units. The logic below will
+        // either return `No` or `ThinLocal`.
+
+        // If processing command line options determined that we're incompatible
+        // with ThinLTO (e.g., `-C lto --emit llvm-ir`) then return that option.
+        if self.opts.cli_forced_thinlto_off {
+            return config::Lto::No;
+        }
+
+        // If `-Z thinlto` specified process that, but note that this is mostly
+        // a deprecated option now that `-C lto=thin` exists.
+        if let Some(enabled) = self.opts.debugging_opts.thinlto {
+            if enabled {
+                return config::Lto::ThinLocal;
+            } else {
+                return config::Lto::No;
+            }
+        }
+
+        // If there's only one codegen unit and LTO isn't enabled then there's
+        // no need for ThinLTO so just return false.
+        if self.codegen_units() == 1 {
+            return config::Lto::No;
+        }
+
+        // Now we're in "defaults" territory. By default we enable ThinLTO for
+        // optimized compiles (anything greater than O0).
+        match self.opts.optimize {
+            config::OptLevel::No => config::Lto::No,
+            _ => config::Lto::ThinLocal,
+        }
+    }
+
+    /// Returns the panic strategy for this compile session. If the user explicitly selected one
+    /// using '-C panic', use that, otherwise use the panic strategy defined by the target.
+    pub fn panic_strategy(&self) -> PanicStrategy {
+        self.opts.cg.panic.unwrap_or(self.target.panic_strategy)
+    }
+    pub fn fewer_names(&self) -> bool {
+        if let Some(fewer_names) = self.opts.debugging_opts.fewer_names {
+            fewer_names
+        } else {
+            let more_names = self.opts.output_types.contains_key(&OutputType::LlvmAssembly)
+                || self.opts.output_types.contains_key(&OutputType::Bitcode)
+                // AddressSanitizer and MemorySanitizer use alloca name when reporting an issue.
+                || self.opts.debugging_opts.sanitizer.intersects(SanitizerSet::ADDRESS | SanitizerSet::MEMORY);
+            !more_names
+        }
+    }
+
+    pub fn unstable_options(&self) -> bool {
+        self.opts.debugging_opts.unstable_options
+    }
+    pub fn is_nightly_build(&self) -> bool {
+        self.opts.unstable_features.is_nightly_build()
+    }
+    pub fn overflow_checks(&self) -> bool {
+        self.opts
+            .cg
+            .overflow_checks
+            .or(self.opts.debugging_opts.force_overflow_checks)
+            .unwrap_or(self.opts.debug_assertions)
+    }
+
+    /// Check whether this compile session and crate type use static crt.
+    pub fn crt_static(&self, crate_type: Option<CrateType>) -> bool {
+        if !self.target.crt_static_respected {
+            // If the target does not opt in to crt-static support, use its default.
+            return self.target.crt_static_default;
+        }
+
+        let requested_features = self.opts.cg.target_feature.split(',');
+        let found_negative = requested_features.clone().any(|r| r == "-crt-static");
+        let found_positive = requested_features.clone().any(|r| r == "+crt-static");
+
+        if found_positive || found_negative {
+            found_positive
+        } else if crate_type == Some(CrateType::ProcMacro)
+            || crate_type == None && self.opts.crate_types.contains(&CrateType::ProcMacro)
+        {
+            // FIXME: When crate_type is not available,
+            // we use compiler options to determine the crate_type.
+            // We can't check `#![crate_type = "proc-macro"]` here.
+            false
+        } else {
+            self.target.crt_static_default
+        }
+    }
+
+    pub fn relocation_model(&self) -> RelocModel {
+        self.opts.cg.relocation_model.unwrap_or(self.target.relocation_model)
+    }
+
+    pub fn code_model(&self) -> Option<CodeModel> {
+        self.opts.cg.code_model.or(self.target.code_model)
+    }
+
+    pub fn tls_model(&self) -> TlsModel {
+        self.opts.debugging_opts.tls_model.unwrap_or(self.target.tls_model)
+    }
+
+    pub fn is_wasi_reactor(&self) -> bool {
+        self.target.options.os == "wasi"
+            && matches!(
+                self.opts.debugging_opts.wasi_exec_model,
+                Some(config::WasiExecModel::Reactor)
+            )
+    }
+
+    pub fn split_debuginfo(&self) -> SplitDebuginfo {
+        self.opts.cg.split_debuginfo.unwrap_or(self.target.split_debuginfo)
+    }
+
+    pub fn target_can_use_split_dwarf(&self) -> bool {
+        !self.target.is_like_windows && !self.target.is_like_osx
+    }
+
+    pub fn must_not_eliminate_frame_pointers(&self) -> bool {
+        // "mcount" function relies on stack pointer.
+        // See <https://sourceware.org/binutils/docs/gprof/Implementation.html>.
+        if self.instrument_mcount() {
+            true
+        } else if let Some(x) = self.opts.cg.force_frame_pointers {
+            x
+        } else {
+            !self.target.eliminate_frame_pointer
+        }
+    }
+
+    pub fn must_emit_unwind_tables(&self) -> bool {
+        // This is used to control the emission of the `uwtable` attribute on
+        // LLVM functions.
+        //
+        // At the very least, unwind tables are needed when compiling with
+        // `-C panic=unwind`.
+        //
+        // On some targets (including windows), however, exceptions include
+        // other events such as illegal instructions, segfaults, etc. This means
+        // that on Windows we end up still needing unwind tables even if the `-C
+        // panic=abort` flag is passed.
+        //
+        // You can also find more info on why Windows needs unwind tables in:
+        //      https://bugzilla.mozilla.org/show_bug.cgi?id=1302078
+        //
+        // If a target requires unwind tables, then they must be emitted.
+        // Otherwise, we can defer to the `-C force-unwind-tables=<yes/no>`
+        // value, if it is provided, or disable them, if not.
+        if self.panic_strategy() == PanicStrategy::Unwind {
+            true
+        } else if self.target.requires_uwtable {
+            true
+        } else {
+            self.opts.cg.force_unwind_tables.unwrap_or(false)
+        }
+    }
+
+    /// Returns the symbol name for the registrar function,
+    /// given the crate `Svh` and the function `DefIndex`.
+    pub fn generate_plugin_registrar_symbol(&self, disambiguator: CrateDisambiguator) -> String {
+        format!("__rustc_plugin_registrar_{}__", disambiguator.to_fingerprint().to_hex())
+    }
+
+    pub fn generate_proc_macro_decls_symbol(&self, disambiguator: CrateDisambiguator) -> String {
+        format!("__rustc_proc_macro_decls_{}__", disambiguator.to_fingerprint().to_hex())
+    }
+
+    pub fn target_filesearch(&self, kind: PathKind) -> filesearch::FileSearch<'_> {
+        filesearch::FileSearch::new(
+            &self.sysroot,
+            self.opts.target_triple.triple(),
+            &self.opts.search_paths,
+            // `target_tlib_path == None` means it's the same as `host_tlib_path`.
+            self.target_tlib_path.as_ref().unwrap_or(&self.host_tlib_path),
+            kind,
+        )
+    }
+    pub fn host_filesearch(&self, kind: PathKind) -> filesearch::FileSearch<'_> {
+        filesearch::FileSearch::new(
+            &self.sysroot,
+            config::host_triple(),
+            &self.opts.search_paths,
+            &self.host_tlib_path,
+            kind,
+        )
+    }
+
+    pub fn set_incr_session_load_dep_graph(&self, load: bool) {
+        let mut incr_comp_session = self.incr_comp_session.borrow_mut();
+
+        if let IncrCompSession::Active { ref mut load_dep_graph, .. } = *incr_comp_session {
+            *load_dep_graph = load;
+        }
+    }
+
+    pub fn incr_session_load_dep_graph(&self) -> bool {
+        let incr_comp_session = self.incr_comp_session.borrow();
+        match *incr_comp_session {
+            IncrCompSession::Active { load_dep_graph, .. } => load_dep_graph,
+            _ => false,
+        }
+    }
+
+    pub fn init_incr_comp_session(
+        &self,
+        session_dir: PathBuf,
+        lock_file: flock::Lock,
+        load_dep_graph: bool,
+    ) {
+        let mut incr_comp_session = self.incr_comp_session.borrow_mut();
+
+        if let IncrCompSession::NotInitialized = *incr_comp_session {
+        } else {
+            panic!("Trying to initialize IncrCompSession `{:?}`", *incr_comp_session)
+        }
+
+        *incr_comp_session =
+            IncrCompSession::Active { session_directory: session_dir, lock_file, load_dep_graph };
+    }
+
+    pub fn finalize_incr_comp_session(&self, new_directory_path: PathBuf) {
+        let mut incr_comp_session = self.incr_comp_session.borrow_mut();
+
+        if let IncrCompSession::Active { .. } = *incr_comp_session {
+        } else {
+            panic!("trying to finalize `IncrCompSession` `{:?}`", *incr_comp_session);
+        }
+
+        // Note: this will also drop the lock file, thus unlocking the directory.
+        *incr_comp_session = IncrCompSession::Finalized { session_directory: new_directory_path };
+    }
+
+    pub fn mark_incr_comp_session_as_invalid(&self) {
+        let mut incr_comp_session = self.incr_comp_session.borrow_mut();
+
+        let session_directory = match *incr_comp_session {
+            IncrCompSession::Active { ref session_directory, .. } => session_directory.clone(),
+            IncrCompSession::InvalidBecauseOfErrors { .. } => return,
+            _ => panic!("trying to invalidate `IncrCompSession` `{:?}`", *incr_comp_session),
+        };
+
+        // Note: this will also drop the lock file, thus unlocking the directory.
+        *incr_comp_session = IncrCompSession::InvalidBecauseOfErrors { session_directory };
+    }
+
+    pub fn incr_comp_session_dir(&self) -> cell::Ref<'_, PathBuf> {
+        let incr_comp_session = self.incr_comp_session.borrow();
+        cell::Ref::map(incr_comp_session, |incr_comp_session| match *incr_comp_session {
+            IncrCompSession::NotInitialized => panic!(
+                "trying to get session directory from `IncrCompSession`: {:?}",
+                *incr_comp_session,
+            ),
+            IncrCompSession::Active { ref session_directory, .. }
+            | IncrCompSession::Finalized { ref session_directory }
+            | IncrCompSession::InvalidBecauseOfErrors { ref session_directory } => {
+                session_directory
+            }
+        })
+    }
+
+    pub fn incr_comp_session_dir_opt(&self) -> Option<cell::Ref<'_, PathBuf>> {
+        self.opts.incremental.as_ref().map(|_| self.incr_comp_session_dir())
+    }
+
+    pub fn print_perf_stats(&self) {
+        println!(
+            "Total time spent computing symbol hashes:      {}",
+            duration_to_secs_str(*self.perf_stats.symbol_hash_time.lock())
+        );
+        println!(
+            "Total queries canonicalized:                   {}",
+            self.perf_stats.queries_canonicalized.load(Ordering::Relaxed)
+        );
+        println!(
+            "normalize_generic_arg_after_erasing_regions:   {}",
+            self.perf_stats.normalize_generic_arg_after_erasing_regions.load(Ordering::Relaxed)
+        );
+        println!(
+            "normalize_projection_ty:                       {}",
+            self.perf_stats.normalize_projection_ty.load(Ordering::Relaxed)
+        );
+    }
+
+    /// We want to know if we're allowed to do an optimization for crate foo from -z fuel=foo=n.
+    /// This expends fuel if applicable, and records fuel if applicable.
+    pub fn consider_optimizing<T: Fn() -> String>(&self, crate_name: &str, msg: T) -> bool {
+        let mut ret = true;
+        if let Some(ref c) = self.optimization_fuel_crate {
+            if c == crate_name {
+                assert_eq!(self.threads(), 1);
+                let mut fuel = self.optimization_fuel.lock();
+                ret = fuel.remaining != 0;
+                if fuel.remaining == 0 && !fuel.out_of_fuel {
+                    self.warn(&format!("optimization-fuel-exhausted: {}", msg()));
+                    fuel.out_of_fuel = true;
+                } else if fuel.remaining > 0 {
+                    fuel.remaining -= 1;
+                }
+            }
+        }
+        if let Some(ref c) = self.print_fuel_crate {
+            if c == crate_name {
+                assert_eq!(self.threads(), 1);
+                self.print_fuel.fetch_add(1, SeqCst);
+            }
+        }
+        ret
+    }
+
+    /// Returns the number of query threads that should be used for this
+    /// compilation
+    pub fn threads(&self) -> usize {
+        self.opts.debugging_opts.threads
+    }
+
+    /// Returns the number of codegen units that should be used for this
+    /// compilation
+    pub fn codegen_units(&self) -> usize {
+        if let Some(n) = self.opts.cli_forced_codegen_units {
+            return n;
+        }
+        if let Some(n) = self.target.default_codegen_units {
+            return n as usize;
+        }
+
+        // If incremental compilation is turned on, we default to a high number
+        // codegen units in order to reduce the "collateral damage" small
+        // changes cause.
+        if self.opts.incremental.is_some() {
+            return 256;
+        }
+
+        // Why is 16 codegen units the default all the time?
+        //
+        // The main reason for enabling multiple codegen units by default is to
+        // leverage the ability for the codegen backend to do codegen and
+        // optimization in parallel. This allows us, especially for large crates, to
+        // make good use of all available resources on the machine once we've
+        // hit that stage of compilation. Large crates especially then often
+        // take a long time in codegen/optimization and this helps us amortize that
+        // cost.
+        //
+        // Note that a high number here doesn't mean that we'll be spawning a
+        // large number of threads in parallel. The backend of rustc contains
+        // global rate limiting through the `jobserver` crate so we'll never
+        // overload the system with too much work, but rather we'll only be
+        // optimizing when we're otherwise cooperating with other instances of
+        // rustc.
+        //
+        // Rather a high number here means that we should be able to keep a lot
+        // of idle cpus busy. By ensuring that no codegen unit takes *too* long
+        // to build we'll be guaranteed that all cpus will finish pretty closely
+        // to one another and we should make relatively optimal use of system
+        // resources
+        //
+        // Note that the main cost of codegen units is that it prevents LLVM
+        // from inlining across codegen units. Users in general don't have a lot
+        // of control over how codegen units are split up so it's our job in the
+        // compiler to ensure that undue performance isn't lost when using
+        // codegen units (aka we can't require everyone to slap `#[inline]` on
+        // everything).
+        //
+        // If we're compiling at `-O0` then the number doesn't really matter too
+        // much because performance doesn't matter and inlining is ok to lose.
+        // In debug mode we just want to try to guarantee that no cpu is stuck
+        // doing work that could otherwise be farmed to others.
+        //
+        // In release mode, however (O1 and above) performance does indeed
+        // matter! To recover the loss in performance due to inlining we'll be
+        // enabling ThinLTO by default (the function for which is just below).
+        // This will ensure that we recover any inlining wins we otherwise lost
+        // through codegen unit partitioning.
+        //
+        // ---
+        //
+        // Ok that's a lot of words but the basic tl;dr; is that we want a high
+        // number here -- but not too high. Additionally we're "safe" to have it
+        // always at the same number at all optimization levels.
+        //
+        // As a result 16 was chosen here! Mostly because it was a power of 2
+        // and most benchmarks agreed it was roughly a local optimum. Not very
+        // scientific.
+        16
+    }
+
+    pub fn teach(&self, code: &DiagnosticId) -> bool {
+        self.opts.debugging_opts.teach && self.diagnostic().must_teach(code)
+    }
+
+    pub fn rust_2015(&self) -> bool {
+        self.opts.edition == Edition::Edition2015
+    }
+
+    /// Are we allowed to use features from the Rust 2018 edition?
+    pub fn rust_2018(&self) -> bool {
+        self.opts.edition >= Edition::Edition2018
+    }
+
+    /// Are we allowed to use features from the Rust 2021 edition?
+    pub fn rust_2021(&self) -> bool {
+        self.opts.edition >= Edition::Edition2021
+    }
+
+    pub fn edition(&self) -> Edition {
+        self.opts.edition
+    }
+
+    /// Returns `true` if we cannot skip the PLT for shared library calls.
+    pub fn needs_plt(&self) -> bool {
+        // Check if the current target usually needs PLT to be enabled.
+        // The user can use the command line flag to override it.
+        let needs_plt = self.target.needs_plt;
+
+        let dbg_opts = &self.opts.debugging_opts;
+
+        let relro_level = dbg_opts.relro_level.unwrap_or(self.target.relro_level);
+
+        // Only enable this optimization by default if full relro is also enabled.
+        // In this case, lazy binding was already unavailable, so nothing is lost.
+        // This also ensures `-Wl,-z,now` is supported by the linker.
+        let full_relro = RelroLevel::Full == relro_level;
+
+        // If user didn't explicitly forced us to use / skip the PLT,
+        // then try to skip it where possible.
+        dbg_opts.plt.unwrap_or(needs_plt || !full_relro)
+    }
+
+    /// Checks if LLVM lifetime markers should be emitted.
+    pub fn emit_lifetime_markers(&self) -> bool {
+        self.opts.optimize != config::OptLevel::No
+        // AddressSanitizer uses lifetimes to detect use after scope bugs.
+        // MemorySanitizer uses lifetimes to detect use of uninitialized stack variables.
+        || self.opts.debugging_opts.sanitizer.intersects(SanitizerSet::ADDRESS | SanitizerSet::MEMORY)
+    }
+
+    pub fn link_dead_code(&self) -> bool {
+        self.opts.cg.link_dead_code.unwrap_or(false)
+    }
+
+    pub fn mark_attr_known(&self, attr: &Attribute) {
+        self.known_attrs.lock().mark(attr)
+    }
+
+    pub fn is_attr_known(&self, attr: &Attribute) -> bool {
+        self.known_attrs.lock().is_marked(attr)
+    }
+
+    pub fn mark_attr_used(&self, attr: &Attribute) {
+        self.used_attrs.lock().mark(attr)
+    }
+
+    pub fn is_attr_used(&self, attr: &Attribute) -> bool {
+        self.used_attrs.lock().is_marked(attr)
+    }
+
+    /// Returns `true` if the attribute's path matches the argument. If it
+    /// matches, then the attribute is marked as used.
+    ///
+    /// This method should only be used by rustc, other tools can use
+    /// `Attribute::has_name` instead, because only rustc is supposed to report
+    /// the `unused_attributes` lint. (`MetaItem` and `NestedMetaItem` are
+    /// produced by lowering an `Attribute` and don't have identity, so they
+    /// only have the `has_name` method, and you need to mark the original
+    /// `Attribute` as used when necessary.)
+    pub fn check_name(&self, attr: &Attribute, name: Symbol) -> bool {
+        let matches = attr.has_name(name);
+        if matches {
+            self.mark_attr_used(attr);
+        }
+        matches
+    }
+
+    pub fn is_proc_macro_attr(&self, attr: &Attribute) -> bool {
+        [sym::proc_macro, sym::proc_macro_attribute, sym::proc_macro_derive]
+            .iter()
+            .any(|kind| self.check_name(attr, *kind))
+    }
+
+    pub fn contains_name(&self, attrs: &[Attribute], name: Symbol) -> bool {
+        attrs.iter().any(|item| self.check_name(item, name))
+    }
+
+    pub fn find_by_name<'a>(
+        &'a self,
+        attrs: &'a [Attribute],
+        name: Symbol,
+    ) -> Option<&'a Attribute> {
+        attrs.iter().find(|attr| self.check_name(attr, name))
+    }
+
+    pub fn filter_by_name<'a>(
+        &'a self,
+        attrs: &'a [Attribute],
+        name: Symbol,
+    ) -> impl Iterator<Item = &'a Attribute> {
+        attrs.iter().filter(move |attr| self.check_name(attr, name))
+    }
+
+    pub fn first_attr_value_str_by_name(
+        &self,
+        attrs: &[Attribute],
+        name: Symbol,
+    ) -> Option<Symbol> {
+        attrs.iter().find(|at| self.check_name(at, name)).and_then(|at| at.value_str())
+    }
+}
+
+fn default_emitter(
+    sopts: &config::Options,
+    registry: rustc_errors::registry::Registry,
+    source_map: Lrc<SourceMap>,
+    emitter_dest: Option<Box<dyn Write + Send>>,
+) -> Box<dyn Emitter + sync::Send> {
+    let macro_backtrace = sopts.debugging_opts.macro_backtrace;
+    match (sopts.error_format, emitter_dest) {
+        (config::ErrorOutputType::HumanReadable(kind), dst) => {
+            let (short, color_config) = kind.unzip();
+
+            if let HumanReadableErrorType::AnnotateSnippet(_) = kind {
+                let emitter =
+                    AnnotateSnippetEmitterWriter::new(Some(source_map), short, macro_backtrace);
+                Box::new(emitter.ui_testing(sopts.debugging_opts.ui_testing))
+            } else {
+                let emitter = match dst {
+                    None => EmitterWriter::stderr(
+                        color_config,
+                        Some(source_map),
+                        short,
+                        sopts.debugging_opts.teach,
+                        sopts.debugging_opts.terminal_width,
+                        macro_backtrace,
+                    ),
+                    Some(dst) => EmitterWriter::new(
+                        dst,
+                        Some(source_map),
+                        short,
+                        false, // no teach messages when writing to a buffer
+                        false, // no colors when writing to a buffer
+                        None,  // no terminal width
+                        macro_backtrace,
+                    ),
+                };
+                Box::new(emitter.ui_testing(sopts.debugging_opts.ui_testing))
+            }
+        }
+        (config::ErrorOutputType::Json { pretty, json_rendered }, None) => Box::new(
+            JsonEmitter::stderr(
+                Some(registry),
+                source_map,
+                pretty,
+                json_rendered,
+                sopts.debugging_opts.terminal_width,
+                macro_backtrace,
+            )
+            .ui_testing(sopts.debugging_opts.ui_testing),
+        ),
+        (config::ErrorOutputType::Json { pretty, json_rendered }, Some(dst)) => Box::new(
+            JsonEmitter::new(
+                dst,
+                Some(registry),
+                source_map,
+                pretty,
+                json_rendered,
+                sopts.debugging_opts.terminal_width,
+                macro_backtrace,
+            )
+            .ui_testing(sopts.debugging_opts.ui_testing),
+        ),
+    }
+}
+
+pub enum DiagnosticOutput {
+    Default,
+    Raw(Box<dyn Write + Send>),
+}
+
+pub fn build_session(
+    sopts: config::Options,
+    local_crate_source_file: Option<PathBuf>,
+    registry: rustc_errors::registry::Registry,
+    diagnostics_output: DiagnosticOutput,
+    driver_lint_caps: FxHashMap<lint::LintId, lint::Level>,
+    file_loader: Option<Box<dyn FileLoader + Send + Sync + 'static>>,
+    target_override: Option<Target>,
+) -> Session {
+    // FIXME: This is not general enough to make the warning lint completely override
+    // normal diagnostic warnings, since the warning lint can also be denied and changed
+    // later via the source code.
+    let warnings_allow = sopts
+        .lint_opts
+        .iter()
+        .filter(|&&(ref key, _)| *key == "warnings")
+        .map(|&(_, ref level)| *level == lint::Allow)
+        .last()
+        .unwrap_or(false);
+    let cap_lints_allow = sopts.lint_cap.map_or(false, |cap| cap == lint::Allow);
+    let can_emit_warnings = !(warnings_allow || cap_lints_allow);
+
+    let write_dest = match diagnostics_output {
+        DiagnosticOutput::Default => None,
+        DiagnosticOutput::Raw(write) => Some(write),
+    };
+
+    let target_cfg = config::build_target_config(&sopts, target_override);
+    let host_triple = TargetTriple::from_triple(config::host_triple());
+    let host = Target::search(&host_triple).unwrap_or_else(|e| {
+        early_error(sopts.error_format, &format!("Error loading host specification: {}", e))
+    });
+
+    let loader = file_loader.unwrap_or_else(|| Box::new(RealFileLoader));
+    let hash_kind = sopts.debugging_opts.src_hash_algorithm.unwrap_or_else(|| {
+        if target_cfg.is_like_msvc {
+            SourceFileHashAlgorithm::Sha1
+        } else {
+            SourceFileHashAlgorithm::Md5
+        }
+    });
+    let source_map = Lrc::new(SourceMap::with_file_loader_and_hash_kind(
+        loader,
+        sopts.file_path_mapping(),
+        hash_kind,
+    ));
+    let emitter = default_emitter(&sopts, registry, source_map.clone(), write_dest);
+
+    let span_diagnostic = rustc_errors::Handler::with_emitter_and_flags(
+        emitter,
+        sopts.debugging_opts.diagnostic_handler_flags(can_emit_warnings),
+    );
+
+    let self_profiler = if let SwitchWithOptPath::Enabled(ref d) = sopts.debugging_opts.self_profile
+    {
+        let directory =
+            if let Some(ref directory) = d { directory } else { std::path::Path::new(".") };
+
+        let profiler = SelfProfiler::new(
+            directory,
+            sopts.crate_name.as_deref(),
+            &sopts.debugging_opts.self_profile_events,
+        );
+        match profiler {
+            Ok(profiler) => Some(Arc::new(profiler)),
+            Err(e) => {
+                early_warn(sopts.error_format, &format!("failed to create profiler: {}", e));
+                None
+            }
+        }
+    } else {
+        None
+    };
+
+    let mut parse_sess = ParseSess::with_span_handler(span_diagnostic, source_map);
+    parse_sess.assume_incomplete_release = sopts.debugging_opts.assume_incomplete_release;
+    let sysroot = match &sopts.maybe_sysroot {
+        Some(sysroot) => sysroot.clone(),
+        None => filesearch::get_or_default_sysroot(),
+    };
+
+    let host_triple = config::host_triple();
+    let target_triple = sopts.target_triple.triple();
+    let host_tlib_path = SearchPath::from_sysroot_and_triple(&sysroot, host_triple);
+    let target_tlib_path = if host_triple == target_triple {
+        None
+    } else {
+        Some(SearchPath::from_sysroot_and_triple(&sysroot, target_triple))
+    };
+
+    let file_path_mapping = sopts.file_path_mapping();
+
+    let local_crate_source_file =
+        local_crate_source_file.map(|path| file_path_mapping.map_prefix(path).0);
+
+    let optimization_fuel_crate = sopts.debugging_opts.fuel.as_ref().map(|i| i.0.clone());
+    let optimization_fuel = Lock::new(OptimizationFuel {
+        remaining: sopts.debugging_opts.fuel.as_ref().map_or(0, |i| i.1),
+        out_of_fuel: false,
+    });
+    let print_fuel_crate = sopts.debugging_opts.print_fuel.clone();
+    let print_fuel = AtomicU64::new(0);
+
+    let working_dir = env::current_dir().unwrap_or_else(|e| {
+        parse_sess.span_diagnostic.fatal(&format!("Current directory is invalid: {}", e)).raise()
+    });
+    let working_dir = file_path_mapping.map_prefix(working_dir);
+
+    let cgu_reuse_tracker = if sopts.debugging_opts.query_dep_graph {
+        CguReuseTracker::new()
+    } else {
+        CguReuseTracker::new_disabled()
+    };
+
+    let prof = SelfProfilerRef::new(
+        self_profiler,
+        sopts.debugging_opts.time_passes || sopts.debugging_opts.time,
+        sopts.debugging_opts.time_passes,
+    );
+
+    let ctfe_backtrace = Lock::new(match env::var("RUSTC_CTFE_BACKTRACE") {
+        Ok(ref val) if val == "immediate" => CtfeBacktrace::Immediate,
+        Ok(ref val) if val != "0" => CtfeBacktrace::Capture,
+        _ => CtfeBacktrace::Disabled,
+    });
+
+    // Try to find a directory containing the Rust `src`, for more details see
+    // the doc comment on the `real_rust_source_base_dir` field.
+    let real_rust_source_base_dir = {
+        // This is the location used by the `rust-src` `rustup` component.
+        let mut candidate = sysroot.join("lib/rustlib/src/rust");
+        if let Ok(metadata) = candidate.symlink_metadata() {
+            // Replace the symlink rustbuild creates, with its destination.
+            // We could try to use `fs::canonicalize` instead, but that might
+            // produce unnecessarily verbose path.
+            if metadata.file_type().is_symlink() {
+                if let Ok(symlink_dest) = std::fs::read_link(&candidate) {
+                    candidate = symlink_dest;
+                }
+            }
+        }
+
+        // Only use this directory if it has a file we can expect to always find.
+        if candidate.join("library/std/src/lib.rs").is_file() { Some(candidate) } else { None }
+    };
+
+    let asm_arch =
+        if target_cfg.allow_asm { InlineAsmArch::from_str(&target_cfg.arch).ok() } else { None };
+
+    let sess = Session {
+        target: target_cfg,
+        host,
+        opts: sopts,
+        host_tlib_path,
+        target_tlib_path,
+        parse_sess,
+        sysroot,
+        local_crate_source_file,
+        working_dir,
+        one_time_diagnostics: Default::default(),
+        crate_types: OnceCell::new(),
+        crate_disambiguator: OnceCell::new(),
+        features: OnceCell::new(),
+        lint_store: OnceCell::new(),
+        recursion_limit: OnceCell::new(),
+        type_length_limit: OnceCell::new(),
+        const_eval_limit: OnceCell::new(),
+        incr_comp_session: OneThread::new(RefCell::new(IncrCompSession::NotInitialized)),
+        cgu_reuse_tracker,
+        prof,
+        perf_stats: PerfStats {
+            symbol_hash_time: Lock::new(Duration::from_secs(0)),
+            queries_canonicalized: AtomicUsize::new(0),
+            normalize_generic_arg_after_erasing_regions: AtomicUsize::new(0),
+            normalize_projection_ty: AtomicUsize::new(0),
+        },
+        code_stats: Default::default(),
+        optimization_fuel_crate,
+        optimization_fuel,
+        print_fuel_crate,
+        print_fuel,
+        jobserver: jobserver::client(),
+        driver_lint_caps,
+        trait_methods_not_found: Lock::new(Default::default()),
+        confused_type_with_std_module: Lock::new(Default::default()),
+        system_library_path: OneThread::new(RefCell::new(Default::default())),
+        ctfe_backtrace,
+        miri_unleashed_features: Lock::new(Default::default()),
+        real_rust_source_base_dir,
+        asm_arch,
+        target_features: FxHashSet::default(),
+        known_attrs: Lock::new(MarkedAttrs::new()),
+        used_attrs: Lock::new(MarkedAttrs::new()),
+        if_let_suggestions: Default::default(),
+    };
+
+    validate_commandline_args_with_session_available(&sess);
+
+    sess
+}
+
+// If it is useful to have a Session available already for validating a
+// commandline argument, you can do so here.
+fn validate_commandline_args_with_session_available(sess: &Session) {
+    // Since we don't know if code in an rlib will be linked to statically or
+    // dynamically downstream, rustc generates `__imp_` symbols that help linkers
+    // on Windows deal with this lack of knowledge (#27438). Unfortunately,
+    // these manually generated symbols confuse LLD when it tries to merge
+    // bitcode during ThinLTO. Therefore we disallow dynamic linking on Windows
+    // when compiling for LLD ThinLTO. This way we can validly just not generate
+    // the `dllimport` attributes and `__imp_` symbols in that case.
+    if sess.opts.cg.linker_plugin_lto.enabled()
+        && sess.opts.cg.prefer_dynamic
+        && sess.target.is_like_windows
+    {
+        sess.err(
+            "Linker plugin based LTO is not supported together with \
+                  `-C prefer-dynamic` when targeting Windows-like targets",
+        );
+    }
+
+    // Make sure that any given profiling data actually exists so LLVM can't
+    // decide to silently skip PGO.
+    if let Some(ref path) = sess.opts.cg.profile_use {
+        if !path.exists() {
+            sess.err(&format!(
+                "File `{}` passed to `-C profile-use` does not exist.",
+                path.display()
+            ));
+        }
+    }
+
+    // Unwind tables cannot be disabled if the target requires them.
+    if let Some(include_uwtables) = sess.opts.cg.force_unwind_tables {
+        if sess.panic_strategy() == PanicStrategy::Unwind && !include_uwtables {
+            sess.err(
+                "panic=unwind requires unwind tables, they cannot be disabled \
+                     with `-C force-unwind-tables=no`.",
+            );
+        }
+
+        if sess.target.requires_uwtable && !include_uwtables {
+            sess.err(
+                "target requires unwind tables, they cannot be disabled with \
+                     `-C force-unwind-tables=no`.",
+            );
+        }
+    }
+
+    // PGO does not work reliably with panic=unwind on Windows. Let's make it
+    // an error to combine the two for now. It always runs into an assertions
+    // if LLVM is built with assertions, but without assertions it sometimes
+    // does not crash and will probably generate a corrupted binary.
+    // We should only display this error if we're actually going to run PGO.
+    // If we're just supposed to print out some data, don't show the error (#61002).
+    if sess.opts.cg.profile_generate.enabled()
+        && sess.target.is_like_msvc
+        && sess.panic_strategy() == PanicStrategy::Unwind
+        && sess.opts.prints.iter().all(|&p| p == PrintRequest::NativeStaticLibs)
+    {
+        sess.err(
+            "Profile-guided optimization does not yet work in conjunction \
+                  with `-Cpanic=unwind` on Windows when targeting MSVC. \
+                  See issue #61002 <https://github.com/rust-lang/rust/issues/61002> \
+                  for more information.",
+        );
+    }
+
+    const ASAN_SUPPORTED_TARGETS: &[&str] = &[
+        "aarch64-apple-darwin",
+        "aarch64-fuchsia",
+        "aarch64-unknown-linux-gnu",
+        "x86_64-apple-darwin",
+        "x86_64-fuchsia",
+        "x86_64-unknown-freebsd",
+        "x86_64-unknown-linux-gnu",
+    ];
+    const LSAN_SUPPORTED_TARGETS: &[&str] = &[
+        "aarch64-apple-darwin",
+        "aarch64-unknown-linux-gnu",
+        "x86_64-apple-darwin",
+        "x86_64-unknown-linux-gnu",
+    ];
+    const MSAN_SUPPORTED_TARGETS: &[&str] =
+        &["aarch64-unknown-linux-gnu", "x86_64-unknown-freebsd", "x86_64-unknown-linux-gnu"];
+    const TSAN_SUPPORTED_TARGETS: &[&str] = &[
+        "aarch64-apple-darwin",
+        "aarch64-unknown-linux-gnu",
+        "x86_64-apple-darwin",
+        "x86_64-unknown-freebsd",
+        "x86_64-unknown-linux-gnu",
+    ];
+
+    // Sanitizers can only be used on some tested platforms.
+    for s in sess.opts.debugging_opts.sanitizer {
+        let supported_targets = match s {
+            SanitizerSet::ADDRESS => ASAN_SUPPORTED_TARGETS,
+            SanitizerSet::LEAK => LSAN_SUPPORTED_TARGETS,
+            SanitizerSet::MEMORY => MSAN_SUPPORTED_TARGETS,
+            SanitizerSet::THREAD => TSAN_SUPPORTED_TARGETS,
+            _ => panic!("unrecognized sanitizer {}", s),
+        };
+        if !supported_targets.contains(&&*sess.opts.target_triple.triple()) {
+            sess.err(&format!(
+                "`-Zsanitizer={}` only works with targets: {}",
+                s,
+                supported_targets.join(", ")
+            ));
+        }
+        let conflicting = sess.opts.debugging_opts.sanitizer - s;
+        if !conflicting.is_empty() {
+            sess.err(&format!(
+                "`-Zsanitizer={}` is incompatible with `-Zsanitizer={}`",
+                s, conflicting,
+            ));
+            // Don't report additional errors.
+            break;
+        }
+    }
+}
+
+/// Holds data on the current incremental compilation session, if there is one.
+#[derive(Debug)]
+pub enum IncrCompSession {
+    /// This is the state the session will be in until the incr. comp. dir is
+    /// needed.
+    NotInitialized,
+    /// This is the state during which the session directory is private and can
+    /// be modified.
+    Active { session_directory: PathBuf, lock_file: flock::Lock, load_dep_graph: bool },
+    /// This is the state after the session directory has been finalized. In this
+    /// state, the contents of the directory must not be modified any more.
+    Finalized { session_directory: PathBuf },
+    /// This is an error state that is reached when some compilation error has
+    /// occurred. It indicates that the contents of the session directory must
+    /// not be used, since they might be invalid.
+    InvalidBecauseOfErrors { session_directory: PathBuf },
+}
+
+pub fn early_error(output: config::ErrorOutputType, msg: &str) -> ! {
+    let emitter: Box<dyn Emitter + sync::Send> = match output {
+        config::ErrorOutputType::HumanReadable(kind) => {
+            let (short, color_config) = kind.unzip();
+            Box::new(EmitterWriter::stderr(color_config, None, short, false, None, false))
+        }
+        config::ErrorOutputType::Json { pretty, json_rendered } => {
+            Box::new(JsonEmitter::basic(pretty, json_rendered, None, false))
+        }
+    };
+    let handler = rustc_errors::Handler::with_emitter(true, None, emitter);
+    handler.struct_fatal(msg).emit();
+    rustc_errors::FatalError.raise();
+}
+
+pub fn early_warn(output: config::ErrorOutputType, msg: &str) {
+    let emitter: Box<dyn Emitter + sync::Send> = match output {
+        config::ErrorOutputType::HumanReadable(kind) => {
+            let (short, color_config) = kind.unzip();
+            Box::new(EmitterWriter::stderr(color_config, None, short, false, None, false))
+        }
+        config::ErrorOutputType::Json { pretty, json_rendered } => {
+            Box::new(JsonEmitter::basic(pretty, json_rendered, None, false))
+        }
+    };
+    let handler = rustc_errors::Handler::with_emitter(true, None, emitter);
+    handler.struct_warn(msg).emit();
+}