In preparation for incremental compilation this commit refactors the lint
handling infrastructure in the compiler to be more "eager" and overall more
incremental-friendly. Many passes of the compiler can emit lints at various
points but before this commit all lints were buffered in a table to be emitted
at the very end of compilation. This commit changes these lints to be emitted
immediately during compilation using pre-calculated lint level-related data
structures.
Linting today is split into two phases, one set of "early" lints run on the
`syntax::ast` and a "late" set of lints run on the HIR. This commit moves the
"early" lints to running as late as possible in compilation, just before HIR
lowering. This notably means that we're catching resolve-related lints just
before HIR lowering. The early linting remains a pass very similar to how it was
before, maintaining context of the current lint level as it walks the tree.
Post-HIR, however, linting is structured as a method on the `TyCtxt` which
transitively executes a query to calculate lint levels. Each request to lint on
a `TyCtxt` will query the entire crate's 'lint level data structure' and then go
from there about whether the lint should be emitted or not.
The query depends on the entire HIR crate but should be very quick to calculate
(just a quick walk of the HIR) and the red-green system should notice that the
lint level data structure rarely changes, and should hopefully preserve
incrementality.
Overall this resulted in a pretty big change to the test suite now that lints
are emitted much earlier in compilation (on-demand vs only at the end). This in
turn necessitated the addition of many `#![allow(warnings)]` directives
throughout the compile-fail test suite and a number of updates to the UI test
suite.
fixed some clippy warnings in compiletest
This is mainly readability stuff. Whenever the `clone_ref` lint asked me to clone the dereferenced object, I removed the `.clone()` instead, relying on the fact that it has worked so far and the immutable borrow ensures that the value won't change.
add `allow_fail` test attribute
This change allows the user to add an `#[allow_fail]` attribute to
tests that will cause the test to compile & run, but if the test fails
it will not cause the entire test run to fail. The test output will
show the failure, but in yellow instead of red, and also indicate that
it was an allowed failure.
Here is an example of the output: http://imgur.com/a/wt7ga
This change allows the user to add an `#[allow_fail]` attribute to
tests that will cause the test to compile & run, but if the test fails
it will not cause the entire test run to fail. The test output will
show the failure, but in yellow instead of red, and also indicate that
it was an allowed failure.
This is used in wasm32-experimental-emscripten to ensure that emscripten
links against the libc bitcode files produced by the wasm LLVM backend,
instead of using fastcomp.
This adds the experimental targets option to configure so it can be used
by the builders and changes the wasm32 Dockerfile accordingly. Instead
of using LLVM from the emsdk, the builder's emscripten tools now uses
the Rust in-tree LLVM, since this is the one built with wasm support.
This commit deletes the in-tree `getopts` crate in favor of the crates.io-based
`getopts` crate. The main difference here is with a new builder-style API, but
otherwise everything else remains relatively standard.
The 'run-pass' header cause a 'ui' test to execute the result. It is used
to test the lint output, at the same time ensure those lints won't cause
the source code to become compile-fail.
12 run-pass/run-pass-fulldeps tests gained the header and are moved to
ui/ui-fulldeps. After this move, no run-pass/run-pass-fulldeps tests should
rely on the compiler's JSON message. This allows us to stop passing
`--error-format json` in run-pass tests, thus fixing #36516.
ci: Update android ndk and sdk
Make install-sdk.sh and install-ndk.sh more generic so future updates can be made directly on Dockerfile. Update ndk to r13b, which will be necessary to make host builds for android (in the future). Update sdk to r25.2.5 (maybe some emulator performance improvement).
Overall goal: reduce the amount of context a mir pass needs so that it
resembles a query.
- The hooks are no longer "threaded down" to the pass, but rather run
automatically from the top-level (we also thread down the current pass
number, so that the files are sorted better).
- The hook now receives a *single* callback, rather than a callback per-MIR.
- The traits are no longer lifetime parameters, which moved to the
methods -- given that we required
`for<'tcx>` objecs, there wasn't much point to that.
- Several passes now store a `String` instead of a `&'l str` (again, no
point).
use diff crate for compile-fail test diagnostics #41474
Hello!
This fixes#41474
We were using a custom implementation to dump the differences between expected and actual outputs of compile-fail tests.
I removed this internal implementation and added `diff` crate as a new dependency to `compile-fail`.
Again, huge thanks to @nikomatsakis for guiding.
travis: Parallelize tests on Android
Currently our slowest test suite on android, run-pass, takes over 5 times longer
than the x86_64 component (~400 -> ~2200s). Typically QEMU emulation does indeed
add overhead, but not 5x for this kind of workload. One of the slowest parts of
the Android process is that *compilation* happens serially. Tests themselves
need to run single-threaded on the emulator (due to how the test harness works)
and this forces the compiles themselves to be single threaded.
Now Travis gives us more than one core per machine, so it'd be much better if we
could take advantage of them! The emulator itself is still fundamentally
single-threaded, but we should see a nice speedup by sending binaries for it to
run much more quickly.
It turns out that we've already got all the toos to do this in-tree. The
qemu-test-{server,client} that are in use for the ARM Linux testing are a
perfect match for the Android emulator. This commit migrates the custom adb
management code in compiletest/rustbuild to the same qemu-test-{server,client}
implementation that ARM Linux uses.
This allows us to lift the parallelism restriction on the compiletest test
suites, namely run-pass. Consequently although we'll still basically run the
tests themselves in single threaded mode we'll be able to compile all of them in
parallel, keeping the pipeline much more full hopefully and using more cores for
the work at hand. Additionally the architecture here should be a bit speedier as
it should have less overhead than adb which is a whole new process on both the
host and the emulator!
Locally on an 8 core machine I've seen the run-pass test suite speed up from
taking nearly an hour to only taking 5 minutes. I don't think we'll see quite a
drastic speedup on Travis but I'm hoping this change can place the Android tests
well below 2 hours instead of just above 2 hours.
Because the client/server here are now repurposed for more than just QEMU,
they've been renamed to `remote-test-{server,client}`.
Note that this PR does not currently modify how debuginfo tests are executed on
Android. While parallelizable it wouldn't be quite as easy, so that's left to
another day. Thankfull that test suite is much smaller than the run-pass test
suite.
Implement a file-path remapping feature in support of debuginfo and reproducible builds
This PR adds the `-Zremap-path-prefix-from`/`-Zremap-path-prefix-to` commandline option pair and is a more general implementation of #41419. As opposed to the previous attempt, this implementation should enable reproducible builds regardless of the working directory of the compiler.
This implementation of the feature is more general in the sense that the re-mapping will affect *all* paths the compiler emits, including the ones in error messages.
r? @alexcrichton
Currently our slowest test suite on android, run-pass, takes over 5 times longer
than the x86_64 component (~400 -> ~2200s). Typically QEMU emulation does indeed
add overhead, but not 5x for this kind of workload. One of the slowest parts of
the Android process is that *compilation* happens serially. Tests themselves
need to run single-threaded on the emulator (due to how the test harness works)
and this forces the compiles themselves to be single threaded.
Now Travis gives us more than one core per machine, so it'd be much better if we
could take advantage of them! The emulator itself is still fundamentally
single-threaded, but we should see a nice speedup by sending binaries for it to
run much more quickly.
It turns out that we've already got all the tools to do this in-tree. The
qemu-test-{server,client} that are in use for the ARM Linux testing are a
perfect match for the Android emulator. This commit migrates the custom adb
management code in compiletest/rustbuild to the same qemu-test-{server,client}
implementation that ARM Linux uses.
This allows us to lift the parallelism restriction on the compiletest test
suites, namely run-pass. Consequently although we'll still basically run the
tests themselves in single threaded mode we'll be able to compile all of them in
parallel, keeping the pipeline much more full and using more cores for the work
at hand. Additionally the architecture here should be a bit speedier as it
should have less overhead than adb which is a whole new process on both the host
and the emulator!
Locally on an 8 core machine I've seen the run-pass test suite speed up from
taking nearly an hour to only taking 6 minutes. I don't think we'll see quite a
drastic speedup on Travis but I'm hoping this change can place the Android tests
well below 2 hours instead of just above 2 hours.
Because the client/server here are now repurposed for more than just QEMU,
they've been renamed to `remote-test-{server,client}`.
Note that this PR does not currently modify how debuginfo tests are executed on
Android. While parallelizable it wouldn't be quite as easy, so that's left to
another day. Thankfully that test suite is much smaller than the run-pass test
suite.
As a final fix I discovered that the ARM and Android test suites were actually
running all library unit tests (e.g. stdtest, coretest, etc) twice. I've
corrected that to only run tests once which should also give a nice boost in
overall cycle time here.
Handle subtyping in inference through obligations
We currently store subtyping relations in the `TypeVariables` structure as a kind of special case. This branch uses normal obligations to propagate subtyping, thus converting our inference variables into normal fallback. It also does a few other things:
- Removes the (unstable, outdated) support for custom type inference fallback.
- It's not clear how we want this to work, but we know that we don't want it to work the way it currently does.
- The existing support was also just getting in my way.
- Fixes#30225, which was caused by the trait caching code pretending type variables were normal unification variables, when indeed they were not (but now are).
There is one fishy part of these changes: when computing the LUB/GLB of a "bivariant" type parameter, I currently return the `a` value. Bivariant type parameters are only allowed in a very particular situation, where the type parameter is only used as an associated type output, like this:
```rust
pub struct Foo<A, B>
where A: Fn() -> B
{
data: A
}
```
In principle, if one had `T=Foo<A, &'a u32>` and `U=Foo<A, &'b u32>` and (e.g.) `A: for<'a> Fn() -> &'a u32`, then I think that computing the LUB of `T` and `U` might do the wrong thing. Probably the right behavior is just to create a fresh type variable. However, that particular example would not compile (because the where-clause is illegal; `'a` does not appear in any input type). I was not able to make an example that *would* compile and demonstrate this shortcoming, and handling the LUB/GLB was mildly inconvenient, so I left it as is. I am considering whether to revisit this or what.
I have started a crater run to test the impact of these changes.
In some cases, we give multiple primary spans, in which case we would
report one `//~` annotation per primary span. That was very confusing
because these things are reported to the user as a single error.
UI tests would be better here.
rustbuild: Update bootstrap compiler
Now that we've also updated cargo's release process this commit also changes the
download location of Cargo from Cargos archives back to the static.r-l.o
archives. This should ensure that the Cargo download is the exact Cargo paired
with the rustc that we release.
Now that we've also updated cargo's release process this commit also changes the
download location of Cargo from Cargos archives back to the static.r-l.o
archives. This should ensure that the Cargo download is the exact Cargo paired
with the rustc that we release.
It looks like the 6.3.0 MinGW comes with a gdb which has issues (#40184) that an
attempted workaround (#40777) does not actually fix (#40835). The original
motivation for upgradin MinGW was to fix build flakiness (#40546) due to newer
builds not exhibiting the same bug, so let's hope that 6.2.0 isn't too far back
in time and still contains the fix we need.
Closes#40835
Remove internal liblog
This commit deletes the internal liblog in favor of the implementation that
lives on crates.io. Similarly it's also setting a convention for adding crates
to the compiler. The main restriction right now is that we want compiler
implementation details to be unreachable from normal Rust code (e.g. requires a
feature), and by default everything in the sysroot is reachable via `extern
crate`.
The proposal here is to require that crates pulled in have these lines in their
`src/lib.rs`:
#![cfg_attr(rustbuild, feature(staged_api, rustc_private))]
#![cfg_attr(rustbuild, unstable(feature = "rustc_private", issue = "27812"))]
This'll mean that by default they're not using these attributes but when
compiled as part of the compiler they do a few things:
* Mark themselves as entirely unstable via the `staged_api` feature and the
`#![unstable]` attribute.
* Allow usage of other unstable crates via `feature(rustc_private)` which is
required if the crate relies on any other crates to compile (other than std).
In debugging #40546 I was able to reproduce locally finally using
the literal toolchain that the bots were using. I reproduced the error maybe 4
in 10 builds. I also have the 6.3.0 toolchain installed through `pacman` which
has yet to have a failed build.
When attempting to reproduce the bug with the toolchain that this commit
switches to I was unable to reproduce anything after a few builds. I have no
idea what the original problem was, but I'm hoping that it was just some random
bug fixed somewhere along the way.
I don't currently know of a technical reason to stick to the 4.9.2 toolchains we
were previously using. Historcal 5.3.* toolchains would cause llvm to segfault
(maybe a miscompile?) but this seems to have been fixed recently. To me if it
passes CI then I think we're good.
Closes#40546
This commit deletes the internal liblog in favor of the implementation that
lives on crates.io. Similarly it's also setting a convention for adding crates
to the compiler. The main restriction right now is that we want compiler
implementation details to be unreachable from normal Rust code (e.g. requires a
feature), and by default everything in the sysroot is reachable via `extern
crate`.
The proposal here is to require that crates pulled in have these lines in their
`src/lib.rs`:
#![cfg_attr(rustbuild, feature(staged_api, rustc_private))]
#![cfg_attr(rustbuild, unstable(feature = "rustc_private", issue = "27812"))]
This'll mean that by default they're not using these attributes but when
compiled as part of the compiler they do a few things:
* Mark themselves as entirely unstable via the `staged_api` feature and the
`#![unstable]` attribute.
* Allow usage of other unstable crates via `feature(rustc_private)` which is
required if the crate relies on any other crates to compile (other than std).
It is more robust to not fail if any directory in a path was created
concurrently. This change lifts rustc internal `create_dir_racy` that
was created to handle such conditions to be new `create_dir_all`
implementation.
