Follow-up on #3833 and #3835. In these PRs, the TB GC was fixed to no
longer cause a stack overflow. One test that motivated it was the test
`fill::horizontal_line` in `tiny_skia`. But not causing stack overflows
was not a large improvents, since it did not fix the fundamental issue:
The tree was too large. The test now ran, but it required gigabytes of
memory and hours of time, whereas it finishes within seconds in Stacked
Borrows.
The problem in that test was that it used [`slice::chunked`](https://doc.rust-lang.org/std/primitive.slice.html#method.chunks) to iterate
a slice in chunks. That iterator is written to reborrow at each call to
`next`, which creates a linear tree with a bunch of intermediary nodes,
which also fragments the `RangeMap` for that allocation.
The solution is to now compact the tree, so that these interior nodes
are removed. Care is taken to not remove nodes that are protected, or
that otherwise restrict their children.
Make TB tree traversal bottom-up
In preparation for #3837, the tree traversal needs to be made bottom-up, because the current top-down tree traversal, coupled with that PR's changes to the garbage collector, can introduce non-deterministic error messages if the GC removes a parent tag of the accessed tag that would have triggered the error first.
This is a breaking change for the diagnostics emitted by TB. The implemented semantics stay the same.
In preparation for #3837, the tree traversal needs to be made bottom-up,
because the current top-down tree traversal, coupled with that PR's
changes to the garbage collector, can introduce non-deterministic error
messages if the GC removes a parent tag of the accessed tag that would
have triggered the error first.
This is a breaking change for the diagnostics emitted by TB. The
implemented semantics stay the same.
Disable tree traversal optimization that is wrong due to lazy nodes.
See #3846 for more information.
For now, the optimization is disabled in a very "hotfix" way, while we think about potential fixes. Nonetheless, this fixes#3846
Rollup of 9 pull requests
Successful merges:
- #129288 (Use subtyping for `UnsafeFnPointer` coercion, too)
- #129405 (Fixing span manipulation and indentation of the suggestion introduced by #126187)
- #129518 (gitignore: ignore ICE reports regardless of directory)
- #129519 (Remove redundant flags from `lower_ty_common` that can be inferred from the HIR)
- #129525 (rustdoc: clean up tuple <-> primitive conversion docs)
- #129526 (Use `FxHasher` on new solver unconditionally)
- #129544 (Removes dead code from the compiler)
- #129553 (add back test for stable-const-can-only-call-stable-const)
- #129590 (Avoid taking reference of &TyKind)
r? `@ghost`
`@rustbot` modify labels: rollup
add back test for stable-const-can-only-call-stable-const
This got accidentally removed in https://github.com/rust-lang/rust/pull/128596 (file `tests/ui/internal/internal-unstable-const.rs`). The test has little to do with "allow internal unstable" though, so add it in a file that already tests various const stability things.
Also tweak the help that suggests to add `rustc_allow_const_fn_unstable` to make it clear that this needs team approval, since it is a fairly big gun.
Use `FxHasher` on new solver unconditionally
r? lqd
This should actually fix the inference problem in ad855fe6db641755c528fb5b1cc0308f3a4bf607, since `HashSet::default` was not inferring the hasher when `HashSet` was coming from the stdlib due to the way that defaulted types/inference vars work. You could cherry-pick this on top of your PR alternatively.
rustdoc: clean up tuple <-> primitive conversion docs
This adds a minor missing feature to `fake_variadic`, so that it can render `impl From<(T,)> for [T; 1]` correctly.
Remove redundant flags from `lower_ty_common` that can be inferred from the HIR
...and then get rid of `lower_ty_common`.
r? ``@fmease`` or re-roll if you're busy!
gitignore: ignore ICE reports regardless of directory
Quite often when working on compiler I end up running into ICEs during the standard library compilation.
These ICEs generate reports in `/library/` and not at the root of the repo, so they aren't `gitignore`d.
I finally ended up committing one today, by accident: https://github.com/rust-lang/rust/pull/129487#event-14002067843
Use subtyping for `UnsafeFnPointer` coercion, too
I overlooked this in #129059, which changed MIR typechecking to use subtyping for other fn pointer coercions.
Fixes#129285
Make `rustc_type_ir` build on stable
This PR fixes a handful of issues that appear in `rustc_type_ir` when trying to build the new solver on stable.
r? ```@compiler-errors```
```@bors``` rollup
handle stage0 `cargo` and `rustc` separately
This change allows setting either `build.cargo` or `build.rustc` without requiring both to be set simultaneously, which was not possible previously.
To try it, set `build.rustc` without setting `build.cargo`, and try to bootstrap on clean build.
Blocker for https://github.com/rust-lang/rust/pull/129152
library: Move unstable API of new_uninit to new features
- `new_zeroed` variants move to `new_zeroed_alloc`
- the `write` fn moves to `box_uninit_write`
The remainder will be stabilized in upcoming patches, as it was decided to only stabilize `uninit*` and `assume_init`.
Build `library/profiler_builtins` from `ci-llvm` if appropriate
Running all of `tests/coverage` requires the LLVM profiler runtime, which requires setting `build.profiler = true`.
Historically, doing that has required checking out the entire `src/llvm-project` submodule. For compiler contributors who otherwise don't need that submodule (thanks to `download-ci-vm`), that's quite inconvenient.
However, thanks to #129116, the downloaded CI LLVM tarball now contains a copy of LLVM's `compiler-rt` directory, which includes all the files needed to build the profiler runtime. So with a little bit of extra logic in bootstrap, we can have `library/profiler_builtins` look for the `compiler-rt` files in `ci-llvm` instead of the `src/llvm-project` submodule.
Rollup of 10 pull requests
Successful merges:
- #128467 (Detect `*` operator on `!Sized` expression)
- #128524 (Don't suggest turning crate-level attributes into outer style)
- #128735 (Add a special case for `CStr`/`CString` in the `improper_ctypes` lint)
- #129429 (Print the generic parameter along with the variance in dumps.)
- #129430 (rustdoc: show exact case-sensitive matches first)
- #129449 (Put Pin::as_deref_mut in impl Pin<Ptr> / rearrange Pin methods)
- #129481 (Update `compiler_builtins` to `0.1.121`)
- #129482 (Add myself to the review rotation for libs)
- #129492 (Make wasm32 platform support docs easier to read)
- #129512 (update the doc comment on lintchecker b/c it parses html now)
r? `@ghost`
`@rustbot` modify labels: rollup
