Auto merge of #62866 - pietroalbini:beta-rollup, r=pietroalbini

[beta] Rollup backports

Cherry picked:

* rustc_target: avoid negative register counts in the SysV x86_64 ABI. #62380
* Fix ICEs when `Self` is used in type aliases #62417
* Raise the default recursion limit to 128 #62450
* Handle errors during error recovery gracefully #62604
* Correctly break out of recovery loop #62607
* Cancel unemitted diagnostics during error recovery #62666
* ci: pin awscli dependencies #62856
* Ensure that checkout is with \n line endings #62564

Rolled up:

* [beta] Backport #62615 #62793
* [beta] Fix #62660 #62792

r? @ghost

.azure-pipelines/steps/install-windows-build-deps.yml

@@ -9,6 +9,9 @@ steps:
   displayName: "Ensure build happens on C:/ instead of D:/"
   condition: and(succeeded(), eq(variables['Agent.OS'], 'Windows_NT'))
 
+- bash: git config --replace-all --global core.autocrlf false
+  displayName: "Disable git automatic line ending conversion (on C:/)"
+
 # Download and install MSYS2, needed primarily for the test suite (run-make) but
 # also used by the MinGW toolchain for assembling things.
 #

.azure-pipelines/steps/run.yml

@@ -12,6 +12,9 @@ steps:
 # Azure's Windows image. Having the conversion enabled caused regressions both
 # in our test suite (it broke miri tests) and in the ecosystem, since we
 # started shipping install scripts with CRLF endings instead of the old LF.
+#
+# Note that we do this a couple times during the build as the PATH and current
+# user/directory change, e.g. when mingw is enabled.
 - bash: git config --global core.autocrlf false
   displayName: "Disable git automatic line ending conversion"
 
@@ -70,6 +73,14 @@ steps:
   displayName: Enable IPv6
   condition: and(succeeded(), eq(variables['Agent.OS'], 'Linux'))
 
+# Disable automatic line ending conversion (again). On Windows, when we're
+# installing dependencies, something switches the git configuration directory or
+# re-enables autocrlf. We've not tracked down the exact cause -- and there may
+# be multiple -- but this should ensure submodules are checked out with the
+# appropriate line endings.
+- bash: git config --replace-all --global core.autocrlf false
+  displayName: "Disable git automatic line ending conversion"
+
 # Check out all our submodules, but more quickly than using git by using one of
 # our custom scripts
 - bash: |
@@ -84,17 +95,37 @@ steps:
   condition: and(succeeded(), eq(variables['Agent.OS'], 'Windows_NT'))
   displayName: Check out submodules (Windows)
 
+# See also the disable for autocrlf above, this just checks that it worked
+#
+# We check both in rust-lang/rust and in a submodule to make sure both are
+# accurate. Submodules are checked out significantly later than the main
+# repository in this script, so settings can (and do!) change between then.
+#
+# Linux (and maybe macOS) builders don't currently have dos2unix so just only
+# run this step on Windows.
+- bash: |
+    set -x
+    # print out the git configuration so we can better investigate failures in
+    # the following
+    git config --list --show-origin
+    dos2unix -ih Cargo.lock src/tools/rust-installer/install-template.sh
+    endings=$(dos2unix -ic Cargo.lock src/tools/rust-installer/install-template.sh)
+    # if endings has non-zero length, error out
+    if [ -n "$endings" ]; then exit 1 ; fi
+  condition: and(succeeded(), eq(variables['Agent.OS'], 'Windows_NT'))
+  displayName: Verify line endings are LF
+
 # Ensure the `aws` CLI is installed so we can deploy later on, cache docker
 # images, etc.
 - bash: |
     set -e
     source src/ci/shared.sh
     sudo apt-get install -y python3-setuptools
-    retry pip3 install awscli --upgrade --user
+    retry pip3 install -r src/ci/awscli-requirements.txt --upgrade --user
     echo "##vso[task.prependpath]$HOME/.local/bin"
   displayName: Install awscli (Linux)
   condition: and(succeeded(), eq(variables['Agent.OS'], 'Linux'))
-- script: pip install awscli
+- script: pip install -r src/ci/awscli-requirements.txt
   displayName: Install awscli (non-Linux)
   condition: and(succeeded(), ne(variables['Agent.OS'], 'Linux'))
 

src/bootstrap/lib.rs

@@ -197,11 +197,11 @@ pub struct Compiler {
 
 #[derive(PartialEq, Eq, Copy, Clone, Debug)]
 pub enum DocTests {
-    // Default, run normal tests and doc tests.
+    /// Run normal tests and doc tests (default).
     Yes,
-    // Do not run any doc tests.
+    /// Do not run any doc tests.
     No,
-    // Only run doc tests.
+    /// Only run doc tests.
     Only,
 }
 
@@ -221,10 +221,10 @@ pub enum GitRepo {
 /// methods specifically on this structure itself (to make it easier to
 /// organize).
 pub struct Build {
-    // User-specified configuration via config.toml
+    /// User-specified configuration from `config.toml`.
     config: Config,
 
-    // Derived properties from the above two configurations
+    // Properties derived from the above configuration
     src: PathBuf,
     out: PathBuf,
     rust_info: channel::GitInfo,
@@ -240,12 +240,12 @@ pub struct Build {
     doc_tests: DocTests,
     verbosity: usize,
 
-    // Targets for which to build.
+    // Targets for which to build
     build: Interned<String>,
     hosts: Vec<Interned<String>>,
     targets: Vec<Interned<String>>,
 
-    // Stage 0 (downloaded) compiler and cargo or their local rust equivalents.
+    // Stage 0 (downloaded) compiler and cargo or their local rust equivalents
     initial_rustc: PathBuf,
     initial_cargo: PathBuf,
 
@@ -255,7 +255,7 @@ pub struct Build {
     cxx: HashMap<Interned<String>, cc::Tool>,
     ar: HashMap<Interned<String>, PathBuf>,
     ranlib: HashMap<Interned<String>, PathBuf>,
-    // Misc
+    // Miscellaneous
     crates: HashMap<Interned<String>, Crate>,
     is_sudo: bool,
     ci_env: CiEnv,

src/ci/awscli-requirements.txt

@@ -0,0 +1,13 @@
+awscli==1.16.201
+botocore==1.12.191
+colorama==0.3.9
+docutils==0.14
+jmespath==0.9.4
+pyasn1==0.4.5
+python-dateutil==2.8.0
+PyYAML==5.1
+rsa==3.4.2
+s3transfer==0.2.1
+six==1.12.0
+urllib3==1.25.3
+futures==3.3.0; python_version < '3.0'

src/librustc/middle/recursion_limit.rs

@@ -12,7 +12,7 @@ use syntax::symbol::{Symbol, sym};
 use rustc_data_structures::sync::Once;
 
 pub fn update_limits(sess: &Session, krate: &ast::Crate) {
-    update_limit(krate, &sess.recursion_limit, sym::recursion_limit, 64);
+    update_limit(krate, &sess.recursion_limit, sym::recursion_limit, 128);
     update_limit(krate, &sess.type_length_limit, sym::type_length_limit, 1048576);
 }
 

src/librustc/session/mod.rs

@@ -9,7 +9,7 @@ use crate::lint;
 use crate::lint::builtin::BuiltinLintDiagnostics;
 use crate::middle::allocator::AllocatorKind;
 use crate::middle::dependency_format;
-use crate::session::config::{OutputType, SwitchWithOptPath};
+use crate::session::config::{OutputType, PrintRequest, SwitchWithOptPath};
 use crate::session::search_paths::{PathKind, SearchPath};
 use crate::util::nodemap::{FxHashMap, FxHashSet};
 use crate::util::common::{duration_to_secs_str, ErrorReported};
@@ -1306,9 +1306,12 @@ fn validate_commandline_args_with_session_available(sess: &Session) {
     // 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.target.options.is_like_msvc &&
-       sess.panic_strategy() == PanicStrategy::Unwind {
+       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 https://github.com/rust-lang/rust/issues/61002 for details.");

src/librustc_resolve/lib.rs

@@ -2519,17 +2519,7 @@ impl<'a> Resolver<'a> {
         debug!("(resolving item) resolving {} ({:?})", name, item.node);
 
         match item.node {
-            ItemKind::Ty(_, ref generics) => {
-                self.with_current_self_item(item, |this| {
-                    this.with_generic_param_rib(HasGenericParams(generics, ItemRibKind), |this| {
-                        let item_def_id = this.definitions.local_def_id(item.id);
-                        this.with_self_rib(Res::SelfTy(Some(item_def_id), None), |this| {
-                            visit::walk_item(this, item)
-                        })
-                    })
-                });
-            }
-
+            ItemKind::Ty(_, ref generics) |
             ItemKind::Existential(_, ref generics) |
             ItemKind::Fn(_, _, ref generics, _) => {
                 self.with_generic_param_rib(

src/librustc_target/abi/call/x86_64.rs

@@ -167,29 +167,44 @@ fn cast_target(cls: &[Option<Class>], size: Size) -> CastTarget {
     target
 }
 
+const MAX_INT_REGS: usize = 6; // RDI, RSI, RDX, RCX, R8, R9
+const MAX_SSE_REGS: usize = 8; // XMM0-7
+
 pub fn compute_abi_info<'a, Ty, C>(cx: &C, fty: &mut FnType<'a, Ty>)
     where Ty: TyLayoutMethods<'a, C> + Copy,
           C: LayoutOf<Ty = Ty, TyLayout = TyLayout<'a, Ty>> + HasDataLayout
 {
-    let mut int_regs = 6; // RDI, RSI, RDX, RCX, R8, R9
-    let mut sse_regs = 8; // XMM0-7
+    let mut int_regs = MAX_INT_REGS;
+    let mut sse_regs = MAX_SSE_REGS;
 
     let mut x86_64_ty = |arg: &mut ArgType<'a, Ty>, is_arg: bool| {
         let mut cls_or_mem = classify_arg(cx, arg);
 
-        let mut needed_int = 0;
-        let mut needed_sse = 0;
         if is_arg {
             if let Ok(cls) = cls_or_mem {
+                let mut needed_int = 0;
+                let mut needed_sse = 0;
                 for &c in &cls {
                     match c {
                         Some(Class::Int) => needed_int += 1,
                         Some(Class::Sse) => needed_sse += 1,
                         _ => {}
                     }
                 }
-                if arg.layout.is_aggregate() && (int_regs < needed_int || sse_regs < needed_sse) {
-                    cls_or_mem = Err(Memory);
+                match (int_regs.checked_sub(needed_int), sse_regs.checked_sub(needed_sse)) {
+                    (Some(left_int), Some(left_sse)) => {
+                        int_regs = left_int;
+                        sse_regs = left_sse;
+                    }
+                    _ => {
+                        // Not enough registers for this argument, so it will be
+                        // passed on the stack, but we only mark aggregates
+                        // explicitly as indirect `byval` arguments, as LLVM will
+                        // automatically put immediates on the stack itself.
+                        if arg.layout.is_aggregate() {
+                            cls_or_mem = Err(Memory);
+                        }
+                    }
                 }
             }
         }
@@ -201,14 +216,14 @@ pub fn compute_abi_info<'a, Ty, C>(cx: &C, fty: &mut FnType<'a, Ty>)
                 } else {
                     // `sret` parameter thus one less integer register available
                     arg.make_indirect();
+                    // NOTE(eddyb) return is handled first, so no registers
+                    // should've been used yet.
+                    assert_eq!(int_regs, MAX_INT_REGS);
                     int_regs -= 1;
                 }
             }
             Ok(ref cls) => {
                 // split into sized chunks passed individually
-                int_regs -= needed_int;
-                sse_regs -= needed_sse;
-
                 if arg.layout.is_aggregate() {
                     let size = arg.layout.size;
                     arg.cast_to(cast_target(cls, size))

src/librustc_typeck/check/mod.rs

@@ -4458,7 +4458,7 @@ pub fn check_bounds_are_used<'tcx>(tcx: TyCtxt<'tcx>, generics: &ty::Generics, t
         return;
     }
 
-    // Make a vector of booleans initially false, set to true when used.
+    // Make a vector of booleans initially `false`; set to `true` when used.
     let mut types_used = vec![false; own_counts.types];
 
     for leaf_ty in ty.walk() {
@@ -4467,7 +4467,7 @@ pub fn check_bounds_are_used<'tcx>(tcx: TyCtxt<'tcx>, generics: &ty::Generics, t
             types_used[index as usize - own_counts.lifetimes] = true;
         } else if let ty::Error = leaf_ty.sty {
             // If there is already another error, do not emit
-            // an error for not using a type Parameter.
+            // an error for not using a type parameter.
             assert!(tcx.sess.has_errors());
             return;
         }

src/libsyntax/parse/parser.rs

@@ -1498,7 +1498,7 @@ impl<'a> Parser<'a> {
         F: Fn(&token::Token) -> bool
     {
         let attrs = self.parse_arg_attributes()?;
-        if let Ok(Some(mut arg)) = self.parse_self_arg() {
+        if let Some(mut arg) = self.parse_self_arg()? {
             arg.attrs = attrs.into();
             return self.recover_bad_self_arg(arg, is_trait_item);
         }
@@ -4675,6 +4675,9 @@ impl<'a> Parser<'a> {
     fn parse_block_tail(&mut self, lo: Span, s: BlockCheckMode) -> PResult<'a, P<Block>> {
         let mut stmts = vec![];
         while !self.eat(&token::CloseDelim(token::Brace)) {
+            if self.token == token::Eof {
+                break;
+            }
             let stmt = match self.parse_full_stmt(false) {
                 Err(mut err) => {
                     err.emit();
@@ -4689,8 +4692,6 @@ impl<'a> Parser<'a> {
             };
             if let Some(stmt) = stmt {
                 stmts.push(stmt);
-            } else if self.token == token::Eof {
-                break;
             } else {
                 // Found only `;` or `}`.
                 continue;
@@ -7454,7 +7455,9 @@ impl<'a> Parser<'a> {
             } else if self.look_ahead(1, |t| *t == token::OpenDelim(token::Paren)) {
                 let ident = self.parse_ident().unwrap();
                 self.bump();  // `(`
-                let kw_name = if let Ok(Some(_)) = self.parse_self_arg_with_attrs() {
+                let kw_name = if let Ok(Some(_)) = self.parse_self_arg_with_attrs()
+                    .map_err(|mut e| e.cancel())
+                {
                     "method"
                 } else {
                     "function"
@@ -7505,7 +7508,9 @@ impl<'a> Parser<'a> {
                 self.eat_to_tokens(&[&token::Gt]);
                 self.bump();  // `>`
                 let (kw, kw_name, ambiguous) = if self.eat(&token::OpenDelim(token::Paren)) {
-                    if let Ok(Some(_)) = self.parse_self_arg_with_attrs() {
+                    if let Ok(Some(_)) = self.parse_self_arg_with_attrs()
+                        .map_err(|mut e| e.cancel())
+                    {
                         ("fn", "method", false)
                     } else {
                         ("fn", "function", false)

src/test/auxiliary/rust_test_helpers.c

@@ -215,6 +215,29 @@ uint64_t get_c_many_params(void *a, void *b, void *c, void *d, struct quad f) {
     return f.c;
 }
 
+struct quad_floats {
+    float a;
+    float b;
+    float c;
+    float d;
+};
+
+float get_c_exhaust_sysv64_ints(
+    void *a,
+    void *b,
+    void *c,
+    void *d,
+    void *e,
+    void *f,
+    // `f` used the last integer register, so `g` goes on the stack.
+    // It also used to bring the "count of available integer registers" down to
+    // `-1` which broke the next SSE-only aggregate argument (`h`) - see #62350.
+    void *g,
+    struct quad_floats h
+) {
+    return h.c;
+}
+
 // Calculates the average of `(x + y) / n` where x: i64, y: f64. There must be exactly n pairs
 // passed as variadic arguments. There are two versions of this function: the
 // variadic one, and the one that takes a `va_list`.