ci: Update dist-{i686,x86_64}-linux to Debian 6
This increases the minimum `{i686,x86_64}-unknown-linux-gnu` platform
from RHEL/CentOS 5 (glibc 2.5 and kernel 2.6.18) to a slightly newer
Debian 6 `squeeze` (glibc 2.11 and kernel 2.6.32). While that release is
already EOL, it happens to match the minimum common versions of two
enterprise distros that do still need Rust support -- RHEL 6 (glibc 2.12
and kernel 2.6.32) and SLES 11 SP4 (glibc 2.11 and kernel 3.0).
Closes#62516.
ci: Set `shell: bash` as a default, remove duplicates
A follow-up to #74406, this commit merely removes the `shell: bash` lines where they were added in favor of setting defaults for *all* "run" steps in the jobs that run the tests.
The changes in #74406 were needed because of an upstream change to the `windows-2019` GitHub Actions image. Previously, the configuration worked fine without specifying `shell: bash`, but for some reason this broke with a new change that was deployed today. The preceding PR was a hotfix to get CI passing, but there was a slightly less duplicative way to specify the default shell for the jobs, which was to set the `defaults.run` option.
This change applies to the `pr`, `try`, `auto`, and `auto-fallible` jobs, which are derived from the YAML-anchor `base-ci-job`. I did not apply these changes to the `master`, `try-success`, `try-failure`, `auto-success`, or `auto-failure` jobs because they have only a few steps.
cc/r? @Mark-Simulacrum
Also, promote defaults.run.shell from inside only the primary jobs to
the top level.
The src/ci/exec-with-shell.py wrapper script was formerly used to change
out the shell mid-job by intercepting a CI_OVERRIDE_SHELL environment
variable. Now, instead, we just set `bash` as the global default across
all jobs, and we also delete the exec-with-shell.py script.
Signed-off-by: Kristofer Rye <kristofer.rye@gmail.com>
This will render the src/ci/exec-with-shell.py script more or less
useless, but we're going to replace that by just using the system bash
instead.
Signed-off-by: Kristofer Rye <kristofer.rye@gmail.com>
Bump version to 1.47
This also bumps to a more recent rustfmt version, just to keep us relatively up to date (though almost nothing has changed in rustfmt we use beyond bumps to the parser infra). No formatting changes as a result of this.
r? @pietroalbini
A follow-up to #74406, this commit merely removes the `shell: bash`
lines where they are explicitly added in favor of setting defaults for
*all* "run" steps.
Signed-off-by: Kristofer Rye <kristofer.rye@gmail.com>
Update cross-compilation README
README seemed rather out of date. I hope the information in my PR is now correct (it was more or less assembled by asking in zulip and learning-by-doing).
RISC-V GNU/Linux as host platform
This PR add a new builder named `dist-riscv64-linux` that builds the compiler toolchain for RISC-V 64-bit GNU/Linux.
r? @alexcrichton
This increases the minimum `{i686,x86_64}-unknown-linux-gnu` platform
from RHEL/CentOS 5 (glibc 2.5 and kernel 2.6.18) to a slightly newer
Debian 6 `squeeze` (glibc 2.11 and kernel 2.6.32). While that release is
already EOL, it happens to match the minimum common versions of two
enterprise distros that do still need Rust support -- RHEL 6 (glibc 2.12
and kernel 2.6.32) and SLES 11 SP4 (glibc 2.11 and kernel 3.0).
build dist for x86_64-unknown-illumos
This change creates a new Docker image, "dist-x86_64-illumos", and sets
things up to build the full set of "dist" packages for illumos hosts, so
that illumos users can use "rustup" to install packages. It also
adjusts the manifest builder to expect complete toolchains for this
platform.
Fix disabled dockerfiles
When the dockerfiles were moved into the host-x86_64 directory, paths
for COPY commands were updated with the new host-x86_64/ prefix. This
suggested that the intended context was src/ci/docker. However, the context
for disabled docker images was src/ci/docker/host-x86_64. This broke the new
paths and prevented src/ci/docker/scripts from being included in the
context at all.
This commit corrects this context allowing docker to find the files it
needs for COPY commands.
Also includes a quick fix to riscv recommended by @bjorn3
In our GitHub Actions setup macOS is too unreliable to gate on it, but
the other builders work fine. This commit splits the macOS builders into
a separate job (called auto-fallible), allowing us to gate on the auto
job without failing due to macOS spurious failures.
This change creates a new Docker image, "dist-x86_64-illumos", and sets
things up to build the full set of "dist" packages for illumos hosts, so
that illumos users can use "rustup" to install packages. It also
adjusts the manifest builder to expect complete toolchains for this
platform.
When the dockerfiles were moved into the host-x86_64 directory, paths
for COPY commands were updated with the new host-x86_64/ prefix. This
suggested that the intended context was src/ci/docker. However, the context
for disabled docker images was src/ci/docker/host-x86_64. This broke the new
paths and prevented src/ci/docker/scripts from being included in the
context at all.
This commit corrects this context allowing docker to find the files it
needs for COPY commands.
Suggested by @bjorn3
Every RUN command creates a new overlay on top of the image as of before
the RUN command. Using fewer RUN commands prevents intermediate overlays
(which in this case would have contained the entire Linux source tree).
We're starting to include native aarch64 machines in our CI, but before
this commit the architecture wasn't included in the cache key for our
Docker images. This means there could be conflicts between images
produced on different architectures, hurting our CI times.
This commit fixes the problem by including the output of `uname -m` in
the cache key.
We need to add runners designed for an aarch64 host system, and it'd be
nice to return an error message if someone tries to run an image
designed for an host architecture in another one.
To start the work on this, this commit moves all the existing builders
in the host-x86_64 directory, and changes the run.sh script to look up
the image in the correct directory based on the host architecture.
Add option for local docker testing.
This adds the option `--dev` to `src/ci/docker/run.sh` so that it will enter an interactive environment for local testing. I have often needed this for testing things, but I always needed to edit this script. I wanted the ability to interact in the environment, run different commands, inspect errors, etc.
Use WASM's saturating casts if they are available
WebAssembly supports saturating floating point to integer casts behind a target feature. The feature is already available on many browsers. Beginning with 1.45 Rust will start defining the behavior of floating point to integer casts to be saturating as well. For this Rust constructs additional checks on top of the `fptoui` / `fptosi` instructions it emits. Here we introduce the possibility for the codegen backend to construct saturating casts itself and only fall back to constructing the checks ourselves if that is not possible.
Resolves part of #73591
Rollup of 10 pull requests
Successful merges:
- #73414 (Implement `slice_strip` feature)
- #73564 (linker: Create GNU_EH_FRAME header by default when producing ELFs)
- #73622 (Deny unsafe ops in unsafe fns in libcore)
- #73684 (add spans to injected coverage counters, extract with CoverageData query)
- #73812 (ast_pretty: Pass some token streams and trees by reference)
- #73853 (Add newline to rustc MultiSpan docs)
- #73883 (Compile rustdoc less often.)
- #73885 (Fix wasm32 being broken due to a NodeJS version bump)
- #73903 (Changes required for rustc/cargo to build for iOS targets)
- #73938 (Optimise fast path of checked_ops with `unlikely`)
Failed merges:
r? @ghost
Fix wasm32 being broken due to a NodeJS version bump
Emscripten's SDK [recently bumped the version of NodeJS they shipped](https://github.com/emscripten-core/emsdk/pull/529), but our Dockerfile for the wasm32 builder hardcoded the version number. This will cause consistent CI failures once the currently cached image is rebuilt (either due to a change or due to the cache expiring).
This PR fixes the problem by finding the latest version of NodeJS in the Emscripten SDK and symlinking it to a "latest" directory, which is then added to the `PATH`.
Remove legacy InnoSetup GUI installer
On Windows the InnoSetup `.exe` installer was superseded by the MSI installer long ago. It's no longer needed.
The `.exe` installer hasn't been linked from the [other installation methods](https://forge.rust-lang.org/infra/other-installation-methods.html#standalone) page in many years. As far as I can tell the intent was always to remove this installer once the MSI proved itself. Though admittedly both installers feel very "legacy" at this point.
Removing this would mean we only maintain one Windows GUI installer and would speed up the distribution phase.
As a result of removing InnoSetup, this closes#24397
Test UI tests for pass=check
I'm going to just compare the builder times since I wasn't able to get this working nicely locally (hit some obscure linker error).
Fixes part of #69823
We do not test cross-compilation here as the PR builder lacks a sufficiently
recent LLVM to cross-compile to 32-bit linux. Once we bump the minimum LLVM
version to LLVM 9, this can use normal 32-bit linux.
Revert "ci: allow gating gha on everything but macOS"
The macOS issue on GHA's side seems to be fixed, so this is not needed anymore.
r? @Mark-Simulacrum
Emscripten's SDK recently bumped the version of NodeJS they shipped, but
our Dockerfile for the wasm32 builder hardcoded the version number. This
will cause consistent CI failures once the currently cached image is
rebuilt (either due to a change or due to the cache expiring).
This commit fixes the problem by finding the latest version of NodeJS in
the Emscripten SDK and symlinking it to a "latest" directory, which is
then added to the PATH.
Update mdbook
This updates mdbook to 0.4. The list of changes can be found at https://github.com/rust-lang/mdBook/blob/master/CHANGELOG.md#mdbook-040. I think the most important one is the change to include fonts with the book instead of using the Google Fonts CDN. This adds a few megabytes of fonts to the docs component. It may be possible to share the fonts across the books, but would take a fair bit of work to make that happen, so I'm not sure if it is necessary.
This also removes mdbook-linkcheck. It is currently not being used, and I don't foresee it going back into use anytime soon. I would prefer not to maintain something that isn't being used, and it removes a very large number of dependencies.
ci: allow gating GHA on everything but macOS
In our GitHub Actions setup macOS is too unreliable to gate on it, but the other builders work fine. This commit splits the macOS builders into a separate job (called `auto-fallible`), allowing us to gate on the auto job without failing due to macOS spurious failures.
cc https://github.com/rust-lang/rust-central-station/issues/848
r? @Mark-Simulacrum
ci: disable alt build during try builds
The alt build is not actually needed often, and it can be added back on a case-by-case basis if a specific PR needs access to it.
This will free up a builder.
r? @Mark-Simulacrum
RISC-V Emulated Testing
Adds a disabled docker image on which to run RISC-V tests. Based on the armhf image.
Test using
```
./src/ci/docker/run.sh riscv64gc-linux
```
cc: @msizanoen1
When we're running with dry_run enabled (i.e. all builds do this initially), we're
guaranteed to save of a toolstate of TestFail for tools that aren't tested. In practice,
we do test tools as well, so for those tools we would initially record them as being
TestPass, and then later on re-record the correct state after actually testing them.
However, this would not work well if the build failed for whatever reason (e.g. panicking
in bootstrap, or as was the case in 73097, clippy failing to test successfully), we would
just go on believing that things passed when they in practice did not.
This commit also adjusts saving toolstate to never record clippy explicitly (otherwise, it
would be recorded when building it); eventually that'll likely move to other tools as well
but not yet. This is deemed simpler than checking everywhere we generically save
toolstate.
We also move clippy out of the "toolstate" no-fail-fast build into a separate x.py
invocation; this should no longer be technically required but provides the nice state of
letting us check toolstate for all tools and only then check clippy (giving full results
on every build).
In our GitHub Actions setup macOS is too unreliable to gate on it, but
the other builders work fine. This commit splits the macOS builders into
a separate job (called auto-fallible), allowing us to gate on the auto
job without failing due to macOS spurious failures.
Remove vestigial CI job msvc-aux.
This CI job isn't really doing anything, so it seems prudent to remove it.
For some history:
* This was introduced in #48809 when the msvc job was split in two to keep it under 2 hours (oh the good old days). At the time, this check-aux job did a bunch of things:
* tidy
* src/test/pretty
* src/test/run-pass/pretty
* src/test/run-fail/pretty
* src/test/run-pass-valgrind/pretty
* src/test/run-pass-fulldeps/pretty
* src/test/run-fail-fulldeps/pretty
* Tidy was removed in #60777.
* run-pass and run-pass-fulldeps moved to UI in #63029
* src/test/pretty removed in #58140
* src/test/run-fail moved to UI in #71185
* run-fail-fulldeps removed in #51285
Over time through attrition, the job was left with one lonely thing: `src/test/run-pass-valgrind/pretty`. And of course, this wasn't actually running the "pretty" tests. The normal `run-pass-valgrind` tests ran, and then when it tried to run in "pretty" mode, all the tests were ignored because compiletest thought nothing had changed (apparently compiletest isn't fingerprinting the mode? Needs more investigation…). `run-pass-valgrind` is already being run as part of `x86_64-msvc-1`, so there's no need to run it here.
I've taken the liberty of removing `src/test/run-pass-valgrind/pretty` as a distinct test. I'm guessing from the other PR's that the pretty tests should now live in `src/test/pretty`, and that the team has moved away from doing pretty tests on other parts of the `src/test` tree.
Update musl to 1.1.24
Release notes since previous version 1.1.22:
## 1.1.23 release notes
### new features:
- riscv64 port
- configure now allows customizing AR and RANLIB vars
- header-level support for new linux features in 5.1
### major internal changes:
- removed extern __syscall; syscall header code is now fully self-contained
### performance:
- new math library implementation for log/exp/pow
- aarch64 dynamic tlsdesc function is streamlined
### compatibility & conformance:
- O_TTY_INIT is now defined
- sys/types.h no longer pollutes namespace with sys/sysmacros.h in any profile
- powerpc asm is now compatible with clang internal assembler
### changes for new POSIX interpretations:
- fgetwc now sets stream error indicator on encoding errors
- fmemopen no longer rejects 0 size
### bugs fixed:
- static TLS for shared libraries was allocated wrong on "Variant I" archs
- crash in dladdr reading through uninitialized pointer on non-match
- sigaltstack wrongly errored out on invalid ss_size when doing SS_DISABLE
- getdents function misbehaved with buffer length larger than INT_MAX
- set*id could deadlock after fork from multithreaded process
### arch-specfic bugs fixed:
- s390x SO_PEERSEC definition was wrong
- passing of 64-bit syscall arguments was broken on microblaze
- posix_fadvise was broken on mips due to missing 7-arg syscall support
- vrregset_t layout and member naming was wrong on powerpc64
## 1.1.24 release notes
### new features:
- GLOB_TILDE extension to glob
- non-stub catgets localization API, using netbsd binary catalog format
- posix_spawn file actions for [f]chdir (extension, pending future standard)
- secure_getenv function (extension)
- copy_file_range syscall wrapper (Linux extension)
- header-level support for new linux features in 5.2
### performance:
- new fast path for lrint (generic C version) on 32-bit archs
### major internal changes:
- functions involving time are overhauled to be time64-ready in 32-bit archs
- x32 uses the new time64 code paths to replace nasty hacks in syscall glue
### compatibility & conformance:
- support for powerpc[64] unaligned relocation types
- powerpc[64] and sh sys/user.h no longer clash with kernel asm/ptrace.h
- select no longer modifies timeout on failure (or at all)
- mips64 stat results are no longer limited to 32-bit time range
- optreset (BSD extension) now has a public declaration
- support for clang inconsistencies in wchar_t type vs some 32-bit archs
- mips r6 syscall asm no longer has invalid lo/hi register clobbers
- vestigial asm declarations of __tls_get_new are removed (broke some tooling)
- riscv64 mcontext_t mismatch glibc's member naming is corrected
### bugs fixed:
- glob failed to match broken symlinks consistently
- invalid use of interposed calloc to allocate initial TLS
- various dlsym symbol resolution logic errors
- semctl with SEM_STAT_ANY didn't work
- pthread_create with explicit scheduling was subject to priority inversion
- pthread_create failure path had data race for thread count
- timer_create with SIGEV_THREAD notification had data race getting timer id
- wide printf family failed to support l modifier for float formats
### arch-specific bugs fixed:
- x87 floating point stack imbalance in math asm (i386-only CVE-2019-14697)
- x32 clock_adjtime, getrusage, wait3, wait4 produced junk (struct mismatches)
- lseek broken on x32 and mipsn32 with large file offsets
- riscv64 atomics weren't compiler barriers
- riscv64 atomics had broken asm constraints (missing earlyclobber flag)
- arm clone() was broken when compiled as thumb if start function returned
- mipsr6 setjmp/longjmp did not preserve fpu register state correctly
Fixes#71099.
This will reduce costs, as well as lays the groundwork for developers to be able
to locally pull the published docker images without needing AWS credentials.
[CI] Use the latest Python available on Windows
This PR changes our Windows CI to always use the latest Python interpreter available in the GHA tool cache instead of hardcoding Python 3.7.6. This is needed because occasionally GitHub bumps the installed version, deleting the previous one.
This fixes the current GHA outage we're having. I fully expect the outage to propagate to Azure Pipelines in the coming days if we don't merge this, as both GHA and Azure use the same underlying image. Once the PR is merged we can re-enabled the double-gating.
r? @Mark-Simulacrum
This commit changes our Windows CI to always use the latest Python
interpreter available in the GHA tool cache instead of hardcoding Python
3.7.6. This is needed because occasionally GitHub bumps the installed
version, deleting the previous one.
Ensure that `./x.py doc --stage 0 src/libstd` works via CI
This was split off from #71645, which recommends that users first try building `libstd` docs with the bootstrap `rustdoc`. This should work in most cases, but will fail if we start using a very recent `rustdoc` feature outside a `#[cfg(not(bootstrap))]`.
It would be very nice to guarantee that `./x.py doc --stage 0 src/libstd` works, since it allows documentation changes to be rendered locally without needing to build the compiler. However, it may put too big a burden on `rustdoc` developers who presumably want to dogfood new features.
Enable "full tools" option on ARM dist builders
This commit switches the `--enable-extended` option on the arm-related
dist builders to `--enable-full-tools`. This alias in `config.py`
corresponds to enabling a few more options:
* `rust.lld = true` - this is the main purpose of this PR, to enable LLD
on ARM-related platforms. This means it will effectively unlock
compilation of wasm programs from an arm host.
* `rust.llvm-tools = true` - it turns out that this option is largely
ignored in rustbuild today. This is only read in one location to set
some flags for the `llvm-tools` package, but the `llvm-tools` package
is already produced on all of these builders. It's predicted that this
will have no effect on build times.
* `rust.codegen-backends = ['llvm']` - historically this also enabled
the emscripten backend, but that has long since been removed.
This brings the ARM dist builders in line with the x86_64 dist builders
using this flag. The hope is that the extra time spent on CI building
LLD will acceptable because it's cached by `sccache`, LLD is a
relatively small C++ project, and the dist builders are all clocking
well under 3 hours (the slowest of all builders) around 2 hours.
There's likely some possible cleanup that can happen with these
configure options since it doesn't look like they've aged too too well,
but I'm hopeful that possible refactorings, if necessary, could be
deferred to future PRs.
This commit switches the `--enable-extended` option on the arm-related
dist builders to `--enable-full-tools`. This alias in `config.py`
corresponds to enabling a few more options:
* `rust.lld = true` - this is the main purpose of this PR, to enable LLD
on ARM-related platforms. This means it will effectively unlock
compilation of wasm programs from an arm host.
* `rust.llvm-tools = true` - it turns out that this option is largely
ignored in rustbuild today. This is only read in one location to set
some flags for the `llvm-tools` package, but the `llvm-tools` package
is already produced on all of these builders. It's predicted that this
will have no effect on build times.
* `rust.codegen-backends = ['llvm']` - historically this also enabled
the emscripten backend, but that has long since been removed.
This brings the ARM dist builders in line with the x86_64 dist builders
using this flag. The hope is that the extra time spent on CI building
LLD will acceptable because it's cached by `sccache`, LLD is a
relatively small C++ project, and the dist builders are all clocking
well under 3 hours (the slowest of all builders) around 2 hours.
There's likely some possible cleanup that can happen with these
configure options since it doesn't look like they've aged too too well,
but I'm hopeful that possible refactorings, if necessary, could be
deferred to future PRs.
ci: run mir-opt tests on PR CI also as 32-bit (for `EMIT_MIR_FOR_EACH_BIT_WIDTH`).
Background: #69916 and [`src/test/mir-opt/README.md`](https://github.com/rust-lang/rust/blob/master/src/test/mir-opt/README.md):
> By default 32 bit and 64 bit targets use the same dump files, which can be problematic in the
presence of pointers in constants or other bit width dependent things. In that case you can add
>
> ```
> // EMIT_MIR_FOR_EACH_BIT_WIDTH
> ```
>
> to your test, causing separate files to be generated for 32bit and 64bit systems.
However, if you change the output of such a test (intentionally or not), or if you add a test and it varies between 32-bit and 64-bit platforms, you have to run this command (for a x64 linux host):
`./x.py test --stage 1 --target x86_64-unknown-linux-gnu --target i686-unknown-linux-gnu --bless src/test/mir-opt`
Otherwise, bors trying to merge the PR will fail, since we test 32-bit targets there.
But we don't on PR CI, which means there's no way the PR author would know (unless they were burnt by this already and know what to look for).
This PR resolves that by running `mir-opt` tests for ~~`i686-unknown-linux-gnu`~~, on PR CI.
**EDIT**: switched to `armv5te-unknown-linux-gnueabi` to work around LLVM 7 crashes (see https://github.com/rust-lang/compiler-builtins/pull/311#issuecomment-612270089), found during testing.
cc @rust-lang/wg-mir-opt @rust-lang/infra
Stop explicitly depending on python 2
This PR revises our previous policy of officially only supporting and testing with python 2 in the CI environment to instead test with python 3. It also changes the defaults to python 3 in our various scripts (usually, by way of `python` rather than `python3` to preserve compatibility with systems that do not have a python 3 available).
The effect of this is that we expect all new patches to support python 3 (and will test as such). We explicitly also expect that patches support python 2.7 as well -- and test as such, though only on one builder. This is intended as a temporary, though likely long-lived, measure to preserve compatibility while looking towards the future which is likely to be a python 3 only world. We do not at this point set a timeline for when we'll drop support for python 2.7; it's plausible that this is months or years into the future, depending on how quickly the ecosystem drops support and how painful it is for us to maintain that support over time.
Closes#65063 (as far as I can tell; please file explicit and separate issues or PRs if not).
Enable rust-lld on dist-x86_64-musl
Add rust-lld to rustup llvm-tools-preview on nightly for musl
I am using a musl distro on my workstation, with `RUSTFLAGS="-C target-feature=-crt-static"` this works fine. I know that `x86_64-unknown-linux-musl` was originally only meant as a target and not as a host. But most problems have been fixed, and I have fewer problems with `unknown` (rustup) than when I am using `x86_64-alpine-linux-musl` (rust installed by the distro). The only thing I am missing is rust-lld in llvm-tools-preview on nightly.
I needed rust-lld for a wasm tutorial. I built rust-lld and tested it with that tutorial, and it worked well. I asked [here](https://users.rust-lang.org/t/enable-rust-lld-on-x86-64-unknown-linux-musl/39851) where to request to enable lld and ended up doing this PR.
I compared llvm-tools-preview `nightly-x86_64-unknown-linux-musl` and `nightly-x86_64-unknown-linux-gnu`: only rust-lld is missing in musl.
I tested the change using:
```bash
./src/ci/docker/run.sh dist-x86_64-musl
```
And I checked that the resulting rust-lld binary runs.
There are some builders that are running out of disk space while
building the Docker images, such as arm-android. This moves and symlinks
/var/lib/docker to the /mnt partition on Linux GHA.
GHA: enable running multiple try builds at the same time
While for auto, try and PR builds we only want the latest commit to be tested, that's not true for try builds: each commit pushed to the branch is a different PR being tested, and we want multiple PRs to be tested in parallel if there is enough demand.
Fixes#70569
While for auto, try and PR builds we only want the latest commit to be
tested, that's not true for try builds: each commit pushed to the branch
is a different PR being tested, and we want multiple PRs to be tested in
parallel if there is enough demand.
Fixes#70569
The /mnt mount point has 53GB of free disk space at the time of writing
this commit, so this moves the build there to avoid running out of disk
space during builds.
On Azure Pipeliones, the C: filesystem is huge with a lot of free space,
while D: is small. By default builds happened in D:, so we added a
script to symlink the big directories to C:, granting us more space.
Filesystem Size Used Avail Use%
C: 256G 143G 114G 56%
D: 14G 2.0G 13G 15%
On GitHub Actions instead C: is almost full, and we have a lot of free
space on D:, where the build happens.
Filesystem Size Used Avail Use%
C: 128G 114G 15G 89%
D: 56G 4.8G 52G 9%
This commit stops creating the symlink on GitHub Actions, fixing the out
of disk space errors we were seeing on some Windows builders.
Apparently the old path we were using for Python 2 on Windows was not
documented, and eventually got removed. This switches our CI to use the
correct path.
Toolstate: remove redundant beta-week check.
I made a bit of a mistake in #69624. The "beta regression" doesn't need to be checked twice.
I also rolled up #69693 to avoid merge conflicts.
This helps us have enough disk space for our builders to be able to complete
successfully. For now, the choices are ad-hoc and 'definitely not needed'. This
should never fail the build, as everything our build needs should be inside
Docker.
Python script PEP8 style guide space formatting and minor Python source cleanup
This PR includes the following changes in the Python sources based on a flake8 3.7.9 (mccabe: 0.6.1, pycodestyle: 2.5.0, pyflakes: 2.1.1) CPython 3.7.6 on Darwin lint:
- PEP8 style guide spacing updates *without* line length changes
- removal of unused local variable assignments in context managers and exception handling
- removal of unused Python import statements
- removal of unnecessary semicolons
Seems like the variable showed by $(ciCheckoutPath) on Azure Pipelines
was wrong, making the toolstate script fail. This commit changes that
function to return the variable previously used by the toolstate script.
Other uses of the function were audited, and there should be no
conflict.
GitHub Actions: preparations, part 2
This PR adds the second batch of commits in preparation for GitHub Actions:
* Removed hardcoded Azure Pipelines variables from `publish_toolstate.sh`
* Fixed a bug in `shared.sh`'s GitHub Actions support
* Fixed binutils missing from MSYS2 on Windows 2019 (GitHub Actions-specific)
* Fixed wrong sysroot in macOS 10.15 onwards (GitHub Actions-specific)
This PR does **not** yet add any builders on GitHub Actions.
r? @alexcrichton
ci: remove 32-bit Apple targets
This PR drops the `i686-apple` and `dist-i686-apple` CI builders, as well as removing the `armv7-apple-ios`, `armv7s-apple-ios` and `i386-apple-ios` targets from the `x86_64-apple` CI builder.
The change was approved in [RFC 2837](https://github.com/rust-lang/rfcs/pull/2837), and it should land in Rust 1.42 stable (so this cycle).
r? @alexcrichton
In their infinite wisdom, Apple decided that (starting from macOS 10.15
onwards) /usr/include is not the location we should all search in for
our beloved C headers. Instead, we should look inside the extremely
intuitive and easily guessable new path:
/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include
Because why not.
When moving the script out of CI configuration and into a proper script
we lost track of the current directory changing (and as such the
parameters of the script needing to be different now).
Toolstate publication only runs if the channel is "nightly" and
previously the toolstate builders did not know that the channel was
nightly (since they are not dist builders).
A look through bootstrap seems to indicate that nothing should directly
depend on the channel being set to `-dev` on the test builders, though
this may cause some problems with UI tests (if for some reason they're
dumping the channel into stderr), but we cannot find evidence of such so
hopefully this is fine.
I spent a while debugging a strage linker error about an outdated `glibc` version, only to discover that it was caused by a stale `obj` directory. It wasn't obviously to be that using the same obj dir with multiple Docker images (for the same target triple) could be a problem.
This commit adds a note to the README, which should hopefully be helpful to anyone else who runs into this issue.
Update the revision of wasi-libc used in wasm32-wasi
This commit updates the `wasi-libc` repository used to build the
wasm32-wasi target's libstd to ensure that both libstd and libc are
using the same wasi snapshot version.
[CI] fix the `! isCI` check in src/ci/run.sh
Using `if [ ! isCI ] || ...` doesn't run any command, just tests `isCI`
as a string, whereas `if ! isCI || ...` will actually run the `isCI`
command and negate its exit status.
This commit updates the `wasi-libc` repository used to build the
wasm32-wasi target's libstd to ensure that both libstd and libc are
using the same wasi snapshot version.
Using `if [ ! isCI ] || ...` doesn't run any command, just tests `isCI`
as a string, whereas `if ! isCI || ...` will actually run the `isCI`
command and negate its exit status.
LLVM 7 is over a year old, which should be plenty for compatibility. The
last LLVM 6 holdout was llvm-emscripten, which went away in #65501.
I've also included a fix for LLVM 8 lacking `MemorySanitizerOptions`,
which was broken by #66522.
Some environment variables (like DEPLOY or DEPLOY_ALT for dist builders,
or IMAGE on Linux builders) are set on a lot of builders, and whether
they should be present or not can be detected automatically based on the
builder name and the platform.
This commit simplifies the CI configuration by automatically setting
those environment variables.
Misc CI improvements
This PR contains some misc improvements to our CI configuration:
* The environment variables for MinGW builders were greatly simplified, with just `CUSTOM_MINGW=1` to tell the install scripts to install the vendored copy. All the others (`MINGW_URL`, `MINGW_DIR`, `MINGW_ARCHIVE` and `MSYS_BITS`) are detected either from the builder name or the environment.
* Collecting CPU stats and running the build were moved into scripts.
* Toolstate scripts validation was previously a separate step, ran just when `IMAGE=mingw-check`. This moves the validation code inside the actual image.
* Vendored copies are now fetched from https://ci-mirrors.rust-lang.org instead of directly from the bucket.
r? @alexcrichton
CentOS 5 only supports SSLv3 without SNI, and to get newer protocols
working we need to download and compile OpenSSL and cURL from our
mirror. Because of that, we can't use the CDN, as CloudFront requires
TLSv1 with SNI.
This commit changes the dist-x86_64-linux image to bypass the CDN for
OpenSSL and cURL.
This commit replaces the mirrors base URL contained in the MINGW_URL
with a CUSTOM_MINGW=1 environment variable. The mirrors base URL will be
fetched instead through the MIRRORS_BASE environment variable, defined
in src/ci/shared.sh.
Currently the `RUST_CONFIGURE_ARGS` variable apparently has a trailing
newline at the end of it due to the way it's configured in yaml. This
causes issues with MSVC's `install-clang.sh` step where the way the bash
syntax works out means that we drop the arg we're trying to add and it
doesn't actually get added!
The hopeful fix here is to tweak how we specify the yaml syntax to not
have a trailing newline, we'll see what CI says about this...
This updates the libc that the `wasm32-wasi` target links against to the
latest revision, mostly just bringing in minor bug fixes and minor wasm
size improvements.
ci: revert msys2 ca-certificates hack
The hack was added because upstream msys2 broke the ca-certificates package, but since then it has been fixed. This reverts CI to use the upstream package.
Part of #65767
Upload toolstates.json to rust-lang-ci2
This PR does two things:
* Following up with https://github.com/rust-lang/rust/pull/65202, it migrates deploying artifacts to CI in a script. Both uploading release artifacts and CPU stats were merged into the same script, designing it to be easily extended.
* Uploads the toolstate JSON to `rust-lang-ci2` along with the release artifacts, both for Linux and Windows. This is needed because @RalfJung wants to stop shipping MIRI when its tests are failing, and the toolstate repo doesn't have entries for each commit. Having the toolstate data (just for that specific commit) on `rust-lang-ci2` will simplify the code a lot.
r? @alexcrichton
cc @RalfJung
Before this commit toolstates.json was stored in /tmp and it wasn't
mounted outside the build container. That caused uploading the file in
the upload-artifacts task to fail, as the file was missing on the host.
Mounting /tmp/toolstates.json alone is not the best approach: if the
file is missing when the container is started the Docker engine will
create a *directory* named /tmp/toolstates.json.
The Docker issue could be solved by pre-creating an empty file named
/tmp/toolstates.json, but doing that could cause problems if bootstrap
fails to generate the file and the toolstate scripts receive an empty
JSON.
The approach I took in this commit is to instead mount a /tmp/toolstate
directory inside Docker, and create the toolstates.json file in it. That
also required a small bootstrap change to ensure the directory is
created if it's missing.
The hack was added because upstream msys2 broke the ca-certificates
package, but since then it has been fixed. This reverts CI to use the
upstream package.
Uploading the toolstate data for each commit will help our release
tooling understand which components are failing, to possibly skip
shipping broken tools to users.
We have a job in our CI (PR's x86_64-gnu-tools) that's supposed to run
only when a submodule is changed in the PR, and it works by having a
task at the start of the build that skips all the following tasks if the
condition isn't met.
Before this commit that task was gated with template parameters, which
is a unique feature of Azure Pipelines. To make our CI more generic this
commit switches the gate to use a simple environment variable plus a
condition, which should be supported on more CI providers.
With #65251 landed there's no need to build two LLVM backends and ship
them with rustc, every target we have now uses the same LLVM backend!
This removes the `src/llvm-emscripten` submodule and additionally
removes all support from rustbuild for building the emscripten LLVM
backend. Multiple codegen backend support is left in place for now, and
this is intended to be an easy 10-15 minute win on CI times by avoiding
having to build LLVM twice.
Upgrade GCC to 8.3.0, glibc to 1.17.0 and crosstool-ng to 1.24.0 for dist-armv7-linux
#62896 was caused by the usage of the GCC 5.2.0 toolchain, which was released back in 2015 and may have bugs affecting LLVM 9.
This PR upgrade GCC to 8.3.0 from 5.2.0, glibc from 1.16.0 to 1.17.0 and crosstool-ng to 1.24.0 only for dist-armv7-linux.
Fixes#62896
r? @alexcrichton
- Compatible with Emscripten 1.38.46-upstream or later upstream.
- Refactors the Emscripten target spec to share code with other wasm
targets.
- Replaces the old incorrect wasm32 C call ABI with the correct one,
preserving the old one as wasm32_bindgen_compat for wasm-bindgen
compatibility.
- Updates the varargs ABI used by Emscripten and deletes the old one.
- Removes the obsolete wasm32-experimental-emscripten target.
- Uses EMCC_CFLAGS on CI to avoid the timeout problems with #63649.
- Refactors the Emscripten target spec to share code with other wasm
targets.
- Replaces the incorrect wasm32 C call ABI with the old asmjs
version, which is correct for both wasm32 and JS.
- Updates the varargs ABI used by Emscripten and deletes the old one.
- Removes the obsolete wasm32-experimental-emscripten target.
- Temporarily makes Emscripten targets use panic=abort by default
because supporting unwinding will require an LLVM patch.
Make all alt builders produce parallel-enabled compilers
We're not quite ready to ship parallel compilers by default, but the alt
builders are not used too much (in theory), so we believe that shipping
a possibly-broken compiler there is not too problematic.
r? @nikomatsakis
We're not quite ready to ship parallel compilers by default, but the alt
builders are not used too much (in theory), so we believe that shipping
a possibly-broken compiler there is not too problematic.
azure: Convert Windows installations scripts to `bash`
Looks like `script`, which uses `cmd.exe`, doesn't have fail-fast
behavior and if a leading command fails the script doesn't actually fail
so long as the last command succeeds. We instead want the opposite
behavior where if any step fails the whole script fails.
I don't really know `cmd.exe` that well, nor powershell, so I've opted
to move everything to `bash` which should be a good common denominator
amongst all platforms to work with. Additionally I know that `set -e`
works to cause scripts to fail fast.
Closes#64551
Looks like `script`, which uses `cmd.exe`, doesn't have fail-fast
behavior and if a leading command fails the script doesn't actually fail
so long as the last command succeeds. We instead want the opposite
behavior where if any step fails the whole script fails.
I don't really know `cmd.exe` that well, nor powershell, so I've opted
to move everything to `bash` which should be a good common denominator
amongst all platforms to work with. Additionally I know that `set -e`
works to cause scripts to fail fast.
Note that some scripts remain as `script` since they don't appear to
work in` bash`. I'm not really sure why but I reorganized them slightly
to have the "meaty command" run at the end.
This commit changes our CI to use two separate IAM users to
authenticate with AWS:
* ci--rust-lang--rust--sccache: has access to the rust-lang-ci-sccache2
S3 bucket and its credentials are available during the whole build.
* ci--rust-lang--rust--upload: has access to the rust-lang-ci2 S3 bucket
and its credentials are available just during the upload step.
The new tokens are available in the `prod-credentials` library.
azure: Disable more LLVM/debug assertions in
This commit disables LLVM/debug assertions in our 5 slowest builders:
* i686-gnu
* i686-gnu-nopt
* i686-msvc-1
* i686-msvc-2
* x86_64-msvc-cargo
This is reducing the amount of test coverage for LLVM/debug assertions,
but we're just unfortunately running out of time on CI too many times.
Some test builds have shown that i686-gnu drops nearly an hour of CI
time by disabling these two assertions. Perhaps when we eventually get
4-core machines we can reenable these, but for now turn them off and
hook them up to the tracking issue at #59637 which will ideally be
repurposes to tracking all of these.
This commit disables LLVM/debug assertions in our 5 slowest builders:
* i686-gnu
* i686-gnu-nopt
* i686-msvc-1
* i686-msvc-2
* x86_64-msvc-cargo
This is reducing the amount of test coverage for LLVM/debug assertions,
but we're just unfortunately running out of time on CI too many times.
Some test builds have shown that i686-gnu drops nearly an hour of CI
time by disabling these two assertions. Perhaps when we eventually get
4-core machines we can reenable these, but for now turn them off and
hook them up to the tracking issue at #59637 which will ideally be
repurposes to tracking all of these.
GitHub only allows people explicitly listed as collaborators on the
repository or who commented on the issue/PR to be assignees, failing to
create the issue if non-assignable people are assigned.
This adds an extra check on CI to make sure all the people listed as
tool maintainers can be assigned to toolstate issues. The check won't be
executed on PR builds due to the lack of a valid token.
ci: move linkcheck from mingw-2 to mingw-1
Running UI tests now takes a huge amount of time on mingw builders
(between 40 and 50 minutes), with mingw-1 builders taking even an hour
less to finish than mingw-2. This PR moves linkcheck from mingw-2 to
mingw-1, removing between 10 and 20 minutes of runtime on the -2
builders.
r? @alexcrichton
Running UI tests now takes a huge amount of time on mingw builders
(between 40 and 50 minutes), with mingw-1 builders taking even an hour
less to finish than mingw-2. This PR moves linkcheck from mingw-2 to
mingw-1, removing between 10 and 20 minutes of runtime on the -2
builders.
