This commit moves the actual code generation in the compiler behind a query
keyed by a codegen unit's name. This ended up entailing quite a few internal
refactorings to enable this, along with a few cut corners:
* The `OutputFilenames` structure is now tracked in the `TyCtxt` as it affects a
whole bunch of trans and such. This is now behind a query and threaded into
the construction of the `TyCtxt`.
* The `TyCtxt` now has a channel "out the back" intended to send data to worker
threads in rustc_trans. This is used as a sort of side effect of the codegen
query but morally what's happening here is the return value of the query
(currently unit but morally a path) is only valid once the background threads
have all finished.
* Dispatching work items to the codegen threads was refactored to only rely on
data in `TyCtxt`, which mostly just involved refactoring where data was
stored, moving it from the translation thread to the controller thread's
`CodegenContext` or the like.
* A new thread locals was introduced in trans to work around the query
system. This is used in the implementation of `assert_module_sources` which
looks like an artifact of the old query system and will presumably go away
once red/green is up and running.
This commit attaches a channel to the LLVM workers to the `TyCtxt` which will
later be used during the codegen query to actually send work to LLVM workers.
Otherwise this commit is just plumbing this channel throughout the compiler to
ensure it reaches the right consumers.
This commit removes the `crate_trans_items` field from the `CrateContext` of
trans. This field, a big map, was calculated during partioning and was a set of
all translation items. This isn't quite incremental-friendly because the map may
change a lot but not have much effect on downstream consumers.
Instead a new query was added for the one location this map was needed, along
with a new comment explaining what the location is doing!
This is a big map that ends up inside of a `CrateContext` during translation for
all codegen units. This means that any change to the map may end up causing an
incremental recompilation of a codegen unit! In order to reduce the amount of
dependencies here between codegen units and the actual input crate this commit
refactors dealing with exported symbols and such into various queries.
The new queries are largely based on existing queries with filled out
implementations for the local crate in addition to external crates, but the main
idea is that while translating codegen untis no unit needs the entire set of
exported symbols, instead they only need queries about particulare `DefId`
instances every now and then.
The linking stage, however, still generates a full list of all exported symbols
from all crates, but that's going to always happen unconditionally anyway, so no
news there!
Otherwise we may emit double errors related to the `#[export_name]` attribute,
for example, and using a query should ensure that it's only emitted at most
once.
This commit moves the `collect_and_partition_translation_items` function into a
query on `TyCtxt` instead of a free function in trans, allowing us to track
dependencies and such of the function.
This commit moves the definition of the `ExportedSymbols` structure to the
`rustc` crate and then creates a query that'll be used to construct the
`ExportedSymbols` set. This in turn uses the reachablity query exposed in the
previous commit.
This commit refactors the the `partitioning::partition` function to operate with
a `TyCtxt` instead of a `SharedCrateContext` in preparation for making it a
query.
This commit refactors the `collect_crate_translation_items` function to only
require the `TyCtxt` instead of a `SharedCrateContext` in preparation for
query-ifying this portion of trans.
Ensure tcp test case passes when disconnected from network
net::tcp::tests::connect_timeout_unroutable fails when the network
is unreachable, like on a laptop disconnected from wifi. Check for
this error and allow the test to pass.
Closes#44645
ci: Upload/download from a new S3 bucket
Moving buckets from us-east-1 to us-west-1 because us-west-1 is where
rust-central-station itself runs and in general is where we have all our other
buckets.
stabilized compiler_fences (fixes#41091)
I did not know what to proceed with "unstable-book" entry. The feature would no longer be unstable so I have deleted it. If it was the wrong call I'll revert it (unfortunately his case is not described in the CONTRIBUTING.md).
stabilized iterator_for_each (closes#42986)
Also updated clippy and rls as these use the iterator_for_each
I've made my first PR's today so most likely I've done something wrong. Sorry about that!
ci: Disable rustc debug assertions on OSX
This commit disables debug assertions for OSX in an effort to improve cycle time
on OSX. It looks like #44610 didn't shave off quite as much time as desired so
let's see how much this helps.
net::tcp::tests::connect_timeout_unroutable fails when the network
is unreachable, like on a laptop disconnected from wifi. Check for
this error and allow the test to pass.
Closes#44645
This commit disables debug assertions for OSX in an effort to improve cycle time
on OSX. It looks like #44610 didn't shave off quite as much time as desired so
let's see how much this helps.
Make use of Travis's conditional jobs.
Conditional jobs: https://docs.travis-ci.com/user/conditional-builds-stages-jobs/#Conditional-Jobs.
Jobs not matching the condition will not be scheduled at all. This allows us to get rid of `$ALLOW_PR`/`$ALLOW_TRY`/`$SKIP_BUILD` in `.travis.yml`, and perfectly prevent spurious PR failures due to flaky macOS machines.
rustbuild: Fix test "test rustdoc" invocation
Previously it would use the librustc output directory which would cause rustdoc
to get entirely recompiled, whereas the intention is that it uses the
already-compiled artifacts from building rustdoc itself, using the tool output
directory
Mention that HashMap::new and HashSet::new do not allocate
The docs for `HashMap::with_capacity` and `HashSet::with_capacity` already say that
> If `capacity` is 0, the hash map/set will not allocate.
However, the docs for `::new` do not say that the initial capacity is 0, and thus promise that a call to `::new` alone does not allocate. This PR fixes that.
rustbuild: Update `cmake` dependency
Should help suppress some warnings from various repos as `cmake` in the newest
version disables warnings by default.
Require +thumb-mode to generate thumb2 code for Android/armv7-a
I am investigating rust's code generation into Gecko by https://bugzilla.mozilla.org/show_bug.cgi?id=1399337.
armv7-linux-androideabi target uses `+v7,+thumb2,+vfp3,+d16,-neon` as target-feature. But `+thumb2` only doesn't generate thumb2 code. To generate thumb2 code, it requires `+thumb-mode`. So we should add it for armv7-linux-androideabi.
r? @alexcrichton
rustc: Preallocate when building the dep graph
This commit alters the `query` function in the dep graph module to preallocate
memory using `with_capacity` instead of relying on automatic growth. Discovered
in #44576 it was found that for the syntex_syntax clean incremental benchmark
the peak memory usage was found when the dep graph was being saved, particularly
the `DepGraphQuery` data structure itself. PRs like #44142 which add more
queries end up just making this much larger!
I didn't see an immediately obvious way to reduce the size of the
`DepGraphQuery` object, but it turns out that `with_capacity` helps quite a bit!
Locally 831 MB was used [before] this commit, and 770 MB is in use at the peak
of the compiler [after] this commit. That's a nice 7.5% improvement! This won't
quite make up for the losses in #44142 but I figured it's a good start.
[before]: https://gist.github.com/alexcrichton/2d2b9c7a65503761925c5a0bcfeb0d1e
[before]: https://gist.github.com/alexcrichton/6da51f2a6184bfb81694cc44f06deb5b
Customize `<FlatMap as Iterator>::fold`
`FlatMap` can use internal iteration for its `fold`, which shows a
performance advantage in the new benchmarks:
test iter::bench_flat_map_chain_ref_sum ... bench: 4,354,111 ns/iter (+/- 108,871)
test iter::bench_flat_map_chain_sum ... bench: 468,167 ns/iter (+/- 2,274)
test iter::bench_flat_map_ref_sum ... bench: 449,616 ns/iter (+/- 6,257)
test iter::bench_flat_map_sum ... bench: 348,010 ns/iter (+/- 1,227)
... where the "ref" benches are using `by_ref()` that isn't optimized.
So this change shows a decent advantage on its own, but much more when
combined with a `chain` iterator that also optimizes `fold`.
travis: Move sccache to the us-west-1 region
Most of the other rust-lang buckets are in us-west-1 and I think the original
bucket was just accidentally created in the us-east-1 region. Let's consolidate
by moving it to the same location as the rest of our buckets.
bring TyCtxt into scope
got comments both from @eddyb and @nikomatsakis (via https://github.com/rust-lang/rust/pull/44505) that we should always put `TyCtxt` in scope
should I just go and import it at other places in the codebase or we just keep doing small improvements?
rustc: Spawn `cmd /c emcc.bat` explicitly
In #42436 the behavior for spawning processes on Windows was tweaked slightly to
fix various bugs, but this caused #42791 as a regression, namely that to spawn
batch scripts they need to be manually spawned with `cmd /c` instead now. This
updates the compiler to handle this case explicitly for Emscripten.
Closes#42791
bump gcc for bootstrap
On Windows, the gcc crate would send /Wall to msvc, which would cause
builds to get flooded with warnings, exploding compile times from one
hour to more than 72! The gcc crate version 0.3.54 changes this behavior
to send /W4 instead, which greatly cuts down on cl.exe flooding the
command prompt window with warnings.
Ipv4Addr and Ipv6Addr convenience constructors.
Introduce convenience constructors for common types.
This introduces the following constructors:
* Ipv6Addr::localhost()
* Ipv6Addr::unspecified()
* Ipv4Addr::localhost()
* Ipv4Addr::unspecified()
The recently added `From` implementations were nice for avoiding the fallibility of conversions from strings like `"127.0.0.1".parse().unwrap()`, and `"::1".parse().unwrap()`, but while the Ipv4 version is roughly comparable in verbosity, the Ipv6 version lacks zero-segment elision, which makes it significantly more awkward: `[0, 0, 0, 0, 0, 0, 0, 0].into()`. While there isn't a clear way to introduce zero elision to type that can infallibly be converted into Ipv6 addresses, this PR resolves the problem for the two most commonly used addresses, which, incidentally, are the ones that suffer the most from the lack of zero-segment elision.
This change is dead simple, and introduces no backwards incompatibility.
See also, [this topic on the inernals board](https://internals.rust-lang.org/t/pre-rfc-convenience-ip-address-constructors/5878)
Implement <Rc<Any>>::downcast
* Implement `<Rc<Any>>::downcast::<T>`
* New unstable method. Works just like Box\<Any\>, but for Rc.
* Any has two cases for its methods: Any and Any + Send; Rc is never Send, so that case is skipped for Rc.
* Motivation for being a method with self is to match Box and there is no user-supplied type; the inner type is Any and downcast does not conflict with any method of Any.
* Arc was skipped because Any itself has no downcast for the case that makes most sense: Any + Send + Sync
