Constrain const vars to error if const types are mismatched
When equating two consts of different types, if either are const variables, constrain them to the correct const error kind.
This helps us avoid "successfully" matching a const against an impl signature but leaving unconstrained const vars, which will lead to incremental ICEs when we call const-eval queries during const projection.
Fixes#109296
The second commit in the stack fixes a regression in the first commit where we end up mentioning `[const error]` in an impl overlap error message. I think the error message changes for the better, but I could implement alternative strategies to avoid this without delaying the overlap error message...
r? `@BoxyUwU`
Drop all messages in bounded channel when destroying the last receiver
Fixes#107466 by splitting the `disconnect` function for receivers/transmitters and dropping all messages in `disconnect_receivers` like the unbounded channel does. Since all receivers must be dropped before the channel is, the messages will already be discarded at that point, so the `Drop` implementation for the channel can be removed.
``@rustbot`` label +T-libs +A-concurrency
Windows: make `Command` prefer non-verbatim paths
When spawning Commands, the path we use can end up being queried using `env::current_exe` (or the equivalent in other languages). Not all applications handle these paths properly therefore we should have a stronger preference for non-verbatim paths when spawning processes.
When running `x.py test` on a downloaded source distribution (e.g.
https://static.rust-lang.org/dist/rustc-<version>-src.tar.gz), the
crates in the vendor directory contain a number of executable files that
cause the tidy test to fail with the following message:
tidy error: binary checked into source: <path>
I see 26 such errors with the 1.68.0 source distribution. A few of these
are .rs source files with incorrect executable permission, but most are
scripts that are correctly marked executable.
This option was added to LLVM in
https://reviews.llvm.org/D121750?id=416339. It makes `llvm_unreachable`
in builds without assertions compile to an `LLVM_BUILTIN_TRAP` instead
of `LLVM_BUILTIN_UNREACHABLE` (which causes undefined behavior and is
equivalent to `std::hint::unreachable_unchecked`).
Having compiler bugs triggering undefined behavior generally seems
undesirable and inconsistent with Rust's goals. There is a check in
`src/tools/tidy/src/style.rs` to reject code using `llvm_unreachable`.
But it is used a lot within LLVM itself.
For instance, this changes a failure I get compiling `libcore` for m68k
from a `SIGSEGV` to `SIGILL`, which seems better though it still doesn't
provide a useful message without switching to an LLVM build with asserts.
It may be best not to do this if it noticeably degrades compiler
performance, but worthwhile if it doesn't do so in any significant way. I
haven't looked into what benchmarks there are for Rustc. That should be
considered before merging.
Include executed tests in the build metrics (and use a custom test display impl)
The main goal of this PR is to include all tests executed in CI inside the build metrics JSON files. I need this for Ferrocene, and `@Mark-Simulacrum` expressed desire to have this as well to ensure all tests are executed at least once somewhere in CI.
Unfortunately implementing this required rewriting inside of bootstrap all of the code to render the test output to console. libtest supports outputting JSON instead of raw text, which we can indeed use to populate the build metrics. Doing that suppresses the console output though, and compared to rustc and Cargo the console output is not included as a JSON field.
Because of that, this PR had to reimplement both the "pretty" format (one test per line, with `rust.verbose-tests = true`), and the "terse" format (the wall of dots, with `rust.verbose-tests = false`). The current implementation should have the exact same output as libtest, except for the benchmark output. libtest's benchmark output is broken in the "terse" format, so since that's our default I slightly improved how it's rendered.
Also, to bring parity with libtest I had to introduce support for coloring output from bootstrap, using the same dependencies `annotate-snippets` uses. It's now possible to use `builder.color_for_stdout(Color::Red, "text")` and `builder.color_for_stderr(Color::Green, "text")` across all of bootstrap, automatically respecting the `--color` flag and whether the stream is a terminal or not.
I recommend reviewing the PR commit-by-commit.
r? `@Mark-Simulacrum`
Rollup of 10 pull requests
Successful merges:
- #106434 (Document `Iterator::sum/product` for Option/Result)
- #108326 (Implement read_buf for a few more types)
- #108842 (Enforce non-lifetime-binders in supertrait preds are not object safe)
- #108896 (new solver: make all goal evaluation able to be automatically rerun )
- #109124 (Add `dist.compression-profile` option to control compression speed)
- #109240 (Walk un-shifted nested `impl Trait` in trait when setting up default trait method assumptions)
- #109385 (fix typo)
- #109386 (add myself to mailmap)
- #109390 (Custom MIR: Support aggregate expressions)
- #109408 (not *all* retags might be explicit in Runtime MIR)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
not *all* retags might be explicit in Runtime MIR
In https://github.com/rust-lang/rust/pull/105317 I made Miri treat `Rvalue::Ref/AddrOf` as implicit retagging sites. This updates the MIR docs accordingly.
For `Rvalue::Ref` I think this makes a lot more sense: creating a new reference is their entire point, so we can avoid bloating the MIR with retags. Also this seems to be the best way to handle cases like `*ptr = &[mut] ...`, where doing a retag is somewhat questionable since maybe `*ptr` points to another place now?
For `Rvalue::AddrOf`, Stacked Borrows needs this because even raw ptrs need some retagging, but Tree Borrows doesn't do ant retagging here and I hope we'll end up with a model where raw pointers don't get retagged.
Custom MIR: Support aggregate expressions
Add support for tuple, array and ADT expressions in custom mir
r? `````@oli-obk````` or `````@tmiasko````` or `````@JakobDegen`````
Walk un-shifted nested `impl Trait` in trait when setting up default trait method assumptions
Fixes a double subtraction in some binder math in return-position `impl Trait` in trait handling code.
Fixes#109239
Add `dist.compression-profile` option to control compression speed
PR #108534 reduced the size of compressed archives, but (as expected) it also resulted in way longer compression times and memory usage during compression.
It's desirable to keep status quo (smaller archives but more CI usage), but it should also be configurable so that downstream users don't have to waste that much time on CI. As a data point, this resulted in doubling the time of Ferrocene's dist jobs, and required us to increase the RAM allocation for one of such jobs.
This PR adds a new `config.toml` setting, `dist.compression-profile`. The values can be:
* `fast`: equivalent to the gzip and xz preset of "1"
* `balanced`: equivalent to the gzip and xz preset of "6" (the CLI defaults as far as I'm aware)
* `best`: equivalent to the gzip present of "9", and our custom xz profile
The default has also been moved back to `balanced`, to try and avoid the compression time regression for downstream users. I don't feel too strongly on the default, and I'm open to changing it.
Also, for the `best` profile the XZ settings do not match the "9" preset used by the CLI, and it might be confusing. Should we create a `custom-rustc-ci`/`ultra` profile for that?
r? ``@Mark-Simulacrum``
new solver: make all goal evaluation able to be automatically rerun
It is generally wrong to call `evaluate_goal` multiple times or `evaluate_goal` and `evaluate_all` for the same `QueryResult` without correctly handling rerunning the goals when inference makes progress. Not doing so will result in the assertion in `evaluate_goal` firing because rerunning the goal will lead to a more accurate `QueryResult`.
Currently there are lots of places that get this wrong and generally it is complex and error prone to handle correctly everywhere. This PR introduces a way to add goals to the `EvalCtxt` and then run all the added goals in a loop so that `evaluate_goal`/`evaluate_all` is not necessary to call manually.
There are a few complications for making everything work "right":
1. the `normalizes-to` hack that replaces the rhs with an unconstrained infer var requires special casing in the new `try_evaluate_added_goals` function similar to how `evaluate_goal`'s assertion special cases that hack.
2. `assemble_candidates_after_normalizing_self_ty`'s normalization step needs to be reran for each candidate otherwise the found candidates will potentially get a more accurate `QueryResult` when rerunning the projection/trait goal which can effect the `QueryResult` of the projection/trait goal.
This is implemented via `EvalCtxt::probe`'s closure's `EvalCtxt` inheriting the added goals of the `EvalCtxt` that `probe` is called on, allowing us to add goals in a probe, and then enter a nested probe for each candidate and evaluate added goals which include the normalization step's goals.
I made `make_canonical_response` evaluate added goals so that it will be hard to mess up the impl of the solver by forgetting to evaluate added goals. Right now the only way to mess this up would be to call `response_no_constraints` (which from the name is obviously weird).
The visibility of `evaluate_goal` means that it can be called from various `compute_x_goal` or candidate assembly functions, this is generally wrong and we should never call `evaluate_goal` manually, instead we should be calling `add_goal`/`add_goals`. This is solved by moving `evaluate_goal` `evaluate_canonical_goal` and `compute_goal` into `eval_ctxt`'s module and making them private so they cannot be called from elsewhere, forcing people to call `add_goal/s` and `evaluate_added_goals_and_make_canonical_resposne`/`try_evaluate_added_goals`
---
Other changes:
- removed the `&& false` that was introduced to the assertion in `evaluate_goal` in #108839
- remove a `!self.did_overflow()` requirement in `search_graph.is_empty()` which causes goals that overflow to ICE
- made `EvalCtxt::eq` take `&mut self` and add all the nested goals via `add_goals` instead of returning them as 99% of call sites just immediately called `EvalCtxt::add_goals` manually.
r? `````@lcnr`````
Implement read_buf for a few more types
Implement read_buf for TcpStream, Stdin, StdinLock, ChildStdout,
ChildStderr (and internally for AnonPipe, Handle, Socket), so
that it skips buffer initialization.
The other provided methods like read_to_string and read_to_end are
implemented in terms of read_buf and so benefit from the optimization
as well.
This commit also implements read_vectored and is_read_vectored where
applicable.
Document `Iterator::sum/product` for Option/Result
Closes#105266
We already document the similar behavior for `collect()` so I believe it makes sense to add this too. The Option/Result implementations *are* documented on their respective pages and the page for `Sum`, but buried amongst many other trait impls which doesn't make it very discoverable.
`````@rustbot````` label +A-docs
Remove the assume(!is_null) from Vec::as_ptr
At a guess, this code is leftover from LLVM was worse at keeping track of the niche information here. In any case, we don't need this anymore: Removing this `assume` doesn't get rid of the `nonnull` attribute on the return type.
Add inlining annotations in `dec2flt`.
Currently, the combination of `dec2flt` being generic and the `FromStr` implementaions
containing inline anttributes causes massive amounts of assembly to be generated whenever
these implementation are used. In addition, the assembly has calls to function which ought to
be inlined, but they are not (even when using lto).
This Pr fixes this.
Distribute libntdll.a with windows-gnu toolchains
This allows the OS loader to load essential functions (e.g. read/write file) at load time instead of lazily doing so at runtime.
r? libs
Split `execute_job` into `execute_job_incr` and `execute_job_non_incr`
`execute_job` was a bit large, so this splits it in 2. Performance was neutral locally, but this may affect bootstrap times.
migrate compiler, bootstrap and compiletest to windows-rs
This PR migrates the compiler, bootstrap, and compiletest to use [windows-rs](https://github.com/microsoft/windows-rs) instead of winapi-rs. windows-rs is the bindings crate provided by Microsoft, and is actively maintained compared to winapi-rs. Not all ecosystem crates have migrated over yet, so there will be a period of time where both crates are used.
windows-rs also provides some nice ergonomics over winapi-rs to convert return values to `Result`s (which found a case where we forgot to check the return value of `CreateFileW`).
