Now that the final bug fixes have been merged into sccache we can start
leveraging sccache on the MSVC builders on AppVeyor instead of relying on the
ad-hoc caching strategy of trigger files and whatnot.
ASan and TSan are supported on macOS, and this commit enables their
support.
The sanitizers are always built as *.dylib on Apple platforms, so they
cannot be statically linked into the corresponding `rustc_?san.rlib`. The
dylibs are directly copied to `lib/rustlib/x86_64-apple-darwin/lib/`
instead.
Note, although Xcode also ships with their own copies of ASan/TSan dylibs,
we cannot use them due to version mismatch.
There is a caveat: the sanitizer libraries are linked as @rpath, so the
user needs to additionally pass `-C rpath`:
rustc -Z sanitizer=address -C rpath file.rs
^~~~~~~~
Otherwise there will be a runtime error:
dyld: Library not loaded: @rpath/libclang_rt.asan_osx_dynamic.dylib
Referenced from: /path/to/executable
Reason: image not found
Abort trap: 6
The next commit includes a temporary change in compiler to force the linker
to emit a usable @rpath.
I've tracked down what I believe is the last spurious sccache failure on #40240
to behavior in mio (carllerche/mio#583), and this commit updates the binaries to
a version which has that fix incorporated.
Previously we would use one builder on Travis to produce two sets of host
compilers for two different targets. Unfortunately though we've recently
increased how much we're building for each target so this is starting to take
unnecessarily long (#40804). This commit splits the dist builders in two by
ensuring that we only dist one target on each builder, which should take a much
shorter amount of time. This should also unblock other work such as landing the
RLS (#40584).
travis: Compile OSX releases with Xcode 7
Unfortunately what we're using right now, Xcode 8.2, cannot compile LLVM for OSX
10.7. We've done this historically and Gecko would like to maintain this
compabitiliby. This commit moves our release builders for OSX to using Xcode 7
which can compile LLVM for 10.7.
The builders running tests continue to use Xcode 8.2, however, because the LLDB
version with Xcode 7, 350, is blacklisted in running our LLDB tests. To continue
running LLDB tests we'll stick with Xcode 8.2.
Unfortunately what we're using right now, Xcode 8.2, cannot compile LLVM for OSX
10.7. We've done this historically and Gecko would like to maintain this
compabitiliby. This commit moves our release builders for OSX to using Xcode 7
which can compile LLVM for 10.7.
The builders running tests continue to use Xcode 8.2, however, because the LLDB
version with Xcode 7, 350, is blacklisted in running our LLDB tests. To continue
running LLDB tests we'll stick with Xcode 8.2.
Attempt to cache git modules
Partial resolution of #40772, appveyor remains to be done once travis looks like it's working ok.
The approach in this PR is based on the `--reference` flag to `git-clone`/`git-submodule --update` and is a compromise based on the current limitations of the tools we're using.
The ideal would be:
1. have a cached pristine copy of rust-lang/rust master in `$HOME/rustsrc` with all submodules initialised
2. clone the PR branch with `git clone --recurse-submodules --reference $HOME/rustsrc git@github.com:rust-lang/rust.git`
This would (in the nonexistent ideal world) use the pristine copy as an object cache for the top level repo and all submodules, transferring over the network only the changes on the branch. Unfortunately, a) there is no way to manually control the initial clone with travis and b) even if there was, cloned submodules don't use the submodules of the reference as an object cache. So the steps we end up with are:
1. have a cached pristine copy of rust-lang/rust master in `$HOME/rustsrc` with all submodules initialised
2. have a cloned PR branch
3. extract the path of each submodule, and explicitly `git submodule update --init --reference $HOME/rustsrc/$module $module` (i.e. point directly to the location of the pristine submodule repo) for each one
I've also taken some care to make this forward compatible, both for adding and removing submodules.
r? @alexcrichton
travis: See if OSX generates crash dumps
I know for a fact we've had sccache segfault on various platforms and we've also
historically had a lot of problems with the linker on OSX. Let's just poke
around in the crash log directory to see if anything exists. If in the future we
see a build we think segfaulted *and* there's contents here then we can add some
bits that actually print out the logs.
There's a suspicion that the OOM killer is killing sccache (maybe) so this adds
some logging to test out that assumption to see if anything dies and is logged
by `dmesg`
I know for a fact we've had sccache segfault on various platforms and we've also
historically had a lot of problems with the linker on OSX. Let's just poke
around in the crash log directory to see if anything exists. If in the future we
see a build we think segfaulted *and* there's contents here then we can add some
bits that actually print out the logs.
When debugging why builds are taking so long it's often useful to get the
timestamp of all log messages as we're not always timing every tiny step of the
build. I wrote a [utility] for prepending a relative timestamp from the start of
a process which is now downloaded to the builders and is what we wrap the entire
build invocation in.
[utility]: https://github.com/alexcrichton/stamp-rsCloses#40577
This commit updates Cargo with rust-lang/cargo#3820 which includes a fix for
rust-lang/cargo#3819. At the same time this also slightly tweaks how rustbuild
builds cargo to ensure that all the build information (including git info and
such) makes its way into the binary.
Closesrust-lang/cargo#3820
travis: Remove compiling OpenSSL through homebrew
I don't believe that we need this any more now that `cargo-vendor` isn't
installed to create a source tarball (that only happens on Linux)
travis: Attempt to debug sccache failures
I can't find anything that'd cause unexpected EOF in the source, so let's try
taking a look at the error logs on failures.
This is a last-ditch attempt to help our pain with dealing with #38878 on the
bots. A new environment variable is added to the compiler,
`RUSTC_RETRY_LINKER_ON_SEGFAULT`, which will instruct the compiler to
automatically retry the final linker invocation if it looks like the linker
segfaulted (up to 2 extra times).
Unfortunately there have been no successful attempts to debug #38878. The only
information seems to be that the linker (e.g. `ld` on OSX) is segfaulting
somewhere in some thread pool implementation. This appears to be spurious as
failed PRs will later merge.
The hope is that this helps the queue keep moving without clogging and delaying
PRs due to #38878.
This change introduces a Dockerfile and script which builds a complete
Fuchsia toolchain which can be used to build Rust distribution for
Fuchsia. We only support cross-compiling at the moment, hence only
setting the target.
Travis only gives us 30GB disk space and we don't currently have an option to
increase that. Each musl target generates "hello world" binaries of about 3.5MB
in size, and we're testing two targets in the same image. We have around 3k
run-pass tests and 2 musl targets which works out to around 20GB. That's
dangerously close to the limit and is causing PRs to bounce.
This PR splits up the builder in two, one for x86_64 musl and the other for
i686. Hopefully that'll keep us under the disk limit.
Closes#40359
PRs can't land againt beta right now because the android bot is filling up on
disk space. I don't really know what's going on but the android bot is the
longest one to run anyway so it'll benefit from being split up regardless.
Now that mozilla/sccache#43 is fixed the caching works for MinGW on Windows. We
still can't use it for MSVC just yet, but I'll try to revive that branch at some
point.
This commit attempts to move more network operations to being retryable through
various operations. For example git submodule updates, downloading snapshots,
etc, are now all in retryable steps.
Hopefully this commit can cut down on the number of network failures we've been
seeing!
Currently CI builds can fail spuriously during the LLVM build (#39003). I
believe this is due to sccache, and I believe that in turn was due to the fact
that the sccache server used to just be a raw mio server. Historically raw mio
servers are quite complicated to get right, but this is why we built Tokio! The
sccache server has been migrated to Tokio which I suspect would fix any latent
issues.
I have no confirmation of this (never been able to reproduce the deadlock
locally), but my hunch is that updating sccache to the master branch will fix
the timeouts during the LLVM build.
The binaries previously came from Gecko's infrastructure, but I've built new
ones by hand for Win/Mac/Linux and uploaded them to our CI bucket.
travis: Add builders without assertions
This commit adds three new builders, one OSX, one Linux, and one MSVC, which
will produce "nightlies" with LLVM assertions disabled. Currently all nightly
releases have LLVM assertions enabled to catch bugs before they reach the
beta/stable channels. The beta/stable channels, however, do not have LLVM
assertions enabled.
Unfortunately though projects like Servo are stuck on nightlies for the near
future at least and are also suffering very long compile times. The purpose of
this commit is to provide artifacts to these projects which are not distributed
through normal channels (e.g. rustup) but are provided for developers to use
locally if need be.
Logistically these builds will all be uploaded to `rustc-builds-alt` instead of
the `rustc-builds` folder of the `rust-lang-ci` bucket. These builds will stay
there forever (until cleaned out if necessary) and there are no plans to
integrate this with rustup and/or the official release process.
This commit adds three new builders, one OSX, one Linux, and one MSVC, which
will produce "nightlies" with LLVM assertions disabled. Currently all nightly
releases have LLVM assertions enabled to catch bugs before they reach the
beta/stable channels. The beta/stable channels, however, do not have LLVM
assertions enabled.
Unfortunately though projects like Servo are stuck on nightlies for the near
future at least and are also suffering very long compile times. The purpose of
this commit is to provide artifacts to these projects which are not distributed
through normal channels (e.g. rustup) but are provided for developers to use
locally if need be.
Logistically these builds will all be uploaded to `rustc-builds-alt` instead of
the `rustc-builds` folder of the `rust-lang-ci` bucket. These builds will stay
there forever (until cleaned out if necessary) and there are no plans to
integrate this with rustup and/or the official release process.
Delete the makefile build system
This PR deletes the makefile build system in favor of the rustbuild build system. The beta has now been branched so 1.16 will continue to be buildable from the makefiles, but going forward 1.17 will only be buildable with rustbuild.
Rustbuild has been the default build system [since 1.15.0](https://github.com/rust-lang/rust/pull/37817) and the makefiles were [proposed for deletion](https://internals.rust-lang.org/t/proposal-for-promoting-rustbuild-to-official-status/4368) at this time back in November of last year.
And now with the deletion of these makefiles we can start getting those sweet sweet improvements of using crates.io crates in the compiler!
Add support for test suites emulated in QEMU
This commit adds support to the build system to execute test suites that cannot
run natively but can instead run inside of a QEMU emulator. A proof-of-concept
builder was added for the `arm-unknown-linux-gnueabihf` target to show off how
this might work.
In general the architecture is to have a server running inside of the emulator
which a local client connects to. The protocol between the server/client
supports compiling tests on the host and running them on the target inside the
emulator.
Closes#33114
travis: Gate on some minimal support for incremental compilation.
This commit adds a travis job that
1. builds a stage2 compiler in incremental mode (but with empty incremental compilation cache), and
2. builds and runs the run-pass test suite also in incremental mode.
Building incrementally with an empty cache makes sure that the compiler doesn't crash in dependency tracking during bootstrapping. Executing the incrementally built test suite gives some measure of confidence that we generate valid code.
Note, however, that the above does not give strong guarantees about the validity of incremental compilation, it just provides a basis for being able to rely on from-scratch incr. comp. builds as reference values in further tests (which then do actual incremental compilation).
r? @alexcrichton
This commit adds support to the build system to execute test suites that cannot
run natively but can instead run inside of a QEMU emulator. A proof-of-concept
builder was added for the `arm-unknown-linux-gnueabihf` target to show off how
this might work.
In general the architecture is to have a server running inside of the emulator
which a local client connects to. The protocol between the server/client
supports compiling tests on the host and running them on the target inside the
emulator.
Closes#33114
travis: Upload all artifacts in build/dist
Previously we only uploaded tarballs, but this modifies Travis/AppVeyor to
upload everything. We shouldn't have anything else in there to worry about and
otherwise we need to be sure to pick up pkg/msi/exe installers.
