Update affected ui & incremental tests to use a user declared variable
bindings instead of temporaries. The former are preserved because of
debuginfo, the latter are not.
Use llvm::computeLTOCacheKey to determine post-ThinLTO CGU reuse
During incremental ThinLTO compilation, we attempt to re-use the
optimized (post-ThinLTO) bitcode file for a module if it is 'safe' to do
so.
Up until now, 'safe' has meant that the set of modules that our current
modules imports from/exports to is unchanged from the previous
compilation session. See PR #67020 and PR #71131 for more details.
However, this turns out be insufficient to guarantee that it's safe
to reuse the post-LTO module (i.e. that optimizing the pre-LTO module
would produce the same result). When LLVM optimizes a module during
ThinLTO, it may look at other information from the 'module index', such
as whether a (non-imported!) global variable is used. If this
information changes between compilation runs, we may end up re-using an
optimized module that (for example) had dead-code elimination run on a
function that is now used by another module.
Fortunately, LLVM implements its own ThinLTO module cache, which is used
when ThinLTO is performed by a linker plugin (e.g. when clang is used to
compile a C proect). Using this cache directly would require extensive
refactoring of our code - but fortunately for us, LLVM provides a
function that does exactly what we need.
The function `llvm::computeLTOCacheKey` is used to compute a SHA-1 hash
from all data that might influence the result of ThinLTO on a module.
In addition to the module imports/exports that we manually track, it
also hashes information about global variables (e.g. their liveness)
which might be used during optimization. By using this function, we
shouldn't have to worry about new LLVM passes breaking our module re-use
behavior.
In LLVM, the output of this function forms part of the filename used to
store the post-ThinLTO module. To keep our current filename structure
intact, this PR just writes out the mapping 'CGU name -> Hash' to a
file. To determine if a post-LTO module should be reused, we compare
hashes from the previous session.
This should unblock PR #75199 - by sheer chance, it seems to have hit
this issue due to the particular CGU partitioning and optimization
decisions that end up getting made.
During incremental ThinLTO compilation, we attempt to re-use the
optimized (post-ThinLTO) bitcode file for a module if it is 'safe' to do
so.
Up until now, 'safe' has meant that the set of modules that our current
modules imports from/exports to is unchanged from the previous
compilation session. See PR #67020 and PR #71131 for more details.
However, this turns out be insufficient to guarantee that it's safe
to reuse the post-LTO module (i.e. that optimizing the pre-LTO module
would produce the same result). When LLVM optimizes a module during
ThinLTO, it may look at other information from the 'module index', such
as whether a (non-imported!) global variable is used. If this
information changes between compilation runs, we may end up re-using an
optimized module that (for example) had dead-code elimination run on a
function that is now used by another module.
Fortunately, LLVM implements its own ThinLTO module cache, which is used
when ThinLTO is performed by a linker plugin (e.g. when clang is used to
compile a C proect). Using this cache directly would require extensive
refactoring of our code - but fortunately for us, LLVM provides a
function that does exactly what we need.
The function `llvm::computeLTOCacheKey` is used to compute a SHA-1 hash
from all data that might influence the result of ThinLTO on a module.
In addition to the module imports/exports that we manually track, it
also hashes information about global variables (e.g. their liveness)
which might be used during optimization. By using this function, we
shouldn't have to worry about new LLVM passes breaking our module re-use
behavior.
In LLVM, the output of this function forms part of the filename used to
store the post-ThinLTO module. To keep our current filename structure
intact, this PR just writes out the mapping 'CGU name -> Hash' to a
file. To determine if a post-LTO module should be reused, we compare
hashes from the previous session.
This should unblock PR #75199 - by sheer chance, it seems to have hit
this issue due to the particular CGU partitioning and optimization
decisions that end up getting made.
If a symbol name can only be imported from one place for a type, and
as long as it was not glob-imported anywhere in the current crate, we
can trim its printed path and print only the name.
This has wide implications on error messages with types, for example,
shortening `std::vec::Vec` to just `Vec`, as long as there is no other
`Vec` importable anywhere.
This adds a new '-Z trim-diagnostic-paths=false' option to control this
feature.
On the good path, with no diagnosis printed, we should try to avoid
issuing this query, so we need to prevent trimmed_def_paths query on
several cases.
This change also relies on a previous commit that differentiates
between `Debug` and `Display` on various rustc types, where the latter
is trimmed and presented to the user and the former is not.
Incomplete features can also be unsound
Some incomplete features do not just ICE, they are also currently unsound (e.g. https://github.com/rust-lang/rust/pull/72029, and also `specialization` -- which is not yet marked incomplete but [should be](https://github.com/rust-lang/rust/pull/71420)). This makes the message reflect that.
While at it I also added a link to the tracking issue, which hopefully should explain what is incomplete/unsound about the feature.
This commit fixes an issue where the codegen backend's selection of LTO
disagreed with what the codegen later thought was being done. Discovered
in #72006 we have a longstanding issue where if `-Clinker-plugin-lto` in
optimized mode is compiled incrementally it will always panic on the
second compilation. The underlying issue turned out to be that the
production of the original artifact determined that LTO should not be
done (because it's being postponed to the linker) but the CGU reuse
selection thought that LTO was done so it was trying to load pre-LTO
artifacts which were never generated.
The fix here is to ensure that the logic when generating code which
determines what kind of LTO is being done is shared amongst the CGU
reuse decision and the backend actually doing LTO. This means that
they'll both be in agreement about whether the previous compilation did
indeed produce incremental pre-LTO artifacts.
Closes#72006
Expansion-driven outline module parsing
After this PR, the parser will not do any conditional compilation or loading of external module files when `mod foo;` is encountered. Instead, the parser only leaves `mod foo;` in place in the AST, with no items filled in. Expansion later kicks in and will load the actual files and do the parsing. This entails that the following is now valid:
```rust
#[cfg(FALSE)]
mod foo {
mod bar {
mod baz; // `foo/bar/baz.rs` doesn't exist, but no error!
}
}
```
Fixes https://github.com/rust-lang/rust/issues/64197.
r? @petrochenkov
Canonicalize inputs to const eval where needed
Canonicalize inputs to const eval, so that they can contain inference variables. Which enables invoking const eval queries even if the current param env has inference variable within it, which can occur during trait selection.
This is a reattempt of #67717, in a far less invasive way.
Fixes#68477
r? @nikomatsakis
cc @eddyb
Unify and improve const-prop lints
Add a single helper method for all lints emitted by const-prop, and make that lint different from the CTFE `const_err` lint. Also consistently check overflow on *arithmetic*, not on the assertion, to make behavior the same for debug and release builds.
See [this summary comment](https://github.com/rust-lang/rust/pull/69185#issuecomment-587924754) for details and the latest status.
In terms of lint formatting, I went for what seems to be the better style: have a general message above the code, and then a specific message at the span:
```
error: this arithmetic operation will overflow
--> $DIR/const-err2.rs:21:18
|
LL | let a_i128 = -std::i128::MIN;
| ^^^^^^^^^^^^^^^ attempt to negate with overflow
```
We could also just have the specific message above and no text at the span if that is preferred.
I also converted some of the existing tests to use compiletest revisions, so that the same test can check a bunch of different compile flags.
Fixes https://github.com/rust-lang/rust/issues/69020.
Helps with https://github.com/rust-lang/rust/issues/69021: debug/release are now consistent, but the assoc-const test in that issue still fails (there is a FIXME in the PR for this). The reason seems to be that const-prop notices the assoc const in `T::N << 42` and does not even bother calling `const_prop` on that operation.
Has no effect on https://github.com/rust-lang/rust/issues/61821; the duplication there has entirely different reasons.
The test checks that we reuse the CGU of a crate when the implementation
details of an `extern crate` have changed.
Signed-off-by: Michal Terepeta <michal.terepeta@gmail.com>
This test does not actually emit any warnings, since
`#![allow(warnings)]` was specified. `compiletest` was erroneously
ignoring `//~` tests and looking only for `//[X]~` ones. As a result of
the changes in the previous commit, we now look for `//~` comments in
incremental tests and expect them to appear in *all* revisions.
Making ICEs and test them in incremental
This adds:
- A way to make the compiler ICE
- A way to check for ICE in `cfail` tests with `should-ice`
- A regression test for issue #65401
I am not sure the attribute added `should-ice` is the best for this job
perf.rlo shows that running the `ConstProp` pass results in
across-the-board wins regardless of debug or opt complilation mode. As a
result, we're turning it on to get the compile time benefits.
`ConstProp` doesn't currently intern the memory used by its `Machine` so
we can't yet propagate allocations which is why
`ConstProp::should_const_prop()` checks if the value being propagated is
a scalar or not.
[mir-opt] Handle return place in ConstProp and improve SimplifyLocals pass
Temporarily rebased on top of #66074. The top 2 commits are new.
r? @oli-obk
