Enable MIR optimizations in test suite again #480

RalfJung · 2018-10-15T06:29:01Z

miri's test suite used to also run with enabled MIR optimizations, but does not any more because of too many bugs. We should get it enabled again, eventually.

For run-pass, see rustup #926 (comment)
For compile-fail, the current blocker is MIR inlining does not handle SourceScopeLocalData rust#66137.

The text was updated successfully, but these errors were encountered:

RalfJung · 2018-10-31T09:29:07Z

The run-pass test suite works with optimizations now, but the compile-fail test suite still explodes with higher optimization levels.

RalfJung · 2019-08-27T06:45:28Z

Had to disable optimizations for run-pass again, likely caused by rust-lang/rust#63580.

wesleywiser · 2019-08-27T13:08:48Z

I think I will have some time to look at this later tonight or tomorrow morning.

wesleywiser · 2019-08-29T00:59:27Z

So, I'm trying to test some local changes to rustc I have and I'm getting build errors when I run ./miri test:

error[E0463]: can't find crate for `core`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `unicode-width`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `term`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `proc_macro`.

To learn more, run the command again with --verbose.
error: `"/home/wesley/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/cargo" "build" "--release" "--manifest-path" "/tmp/xargo.uuhYMKXriUiG/Cargo.toml" "--target" "x86_64-unknown-linux-gnu" "-p" "test"` failed with exit code: Some(101)
note: run with `RUST_BACKTRACE=1` for a backtrace
fatal error: Failed to run xargo

I've built my local rustc with ./x.py build --stage 2, I've added it to rustup as a custom toolchain, and I've run rustup override in the miri repo. Any ideas what might be causing that?

RalfJung · 2019-08-29T06:19:14Z

That is strange. I regularly build Miri against a local rustc. Usually I do ./x.py build src/rustc, maybe that makes the difference? The directory you added as custom toolchain is .../build/x86_64-unknown-linux-gnu/stage2 (maybe with a different target triple), right? Which OS/platform are you on?

In which part of the Miri build process does the error occur? Is it while building Miri, or while building the custom libstd? Please post a complete build log, starting with your ./miri invocation.

wesleywiser · 2019-08-29T12:16:28Z

Ok, so here's what I just tried:

In ~/code/rust/rust run ./x.py build src/rustc
In ~/code/rust/miri run rustup override set stage2 which links to ~/code/rust/rust/build/x86_64-unknown-linux-gnu/stage2
In ~/code/rust/miri run ./miri test

Which does this:

wesley@endurance:~/code/rust/miri> ./miri test
   Compiling autocfg v0.1.6
   Compiling proc-macro2 v0.4.30
   Compiling unicode-xid v0.1.0
   Compiling libc v0.2.62
   Compiling cc v1.0.41
   Compiling syn v0.15.44
   Compiling proc-macro2 v1.0.1
   Compiling unicode-xid v0.2.0
   Compiling failure_derive v0.1.5
   Compiling syn v1.0.5
   Compiling rustc-demangle v0.1.16
   Compiling memchr v2.2.1
   Compiling bitflags v1.1.0
   Compiling cfg-if v0.1.9
   Compiling serde v1.0.99
   Compiling ryu v1.0.0
   Compiling log v0.4.8
   Compiling rand_core v0.4.2
   Compiling lazy_static v1.4.0
   Compiling getrandom v0.1.11
   Compiling byteorder v1.3.2
   Compiling itoa v0.4.4
   Compiling regex-syntax v0.6.11
   Compiling quick-error v1.2.2
   Compiling semver-parser v0.7.0
   Compiling termcolor v1.0.5
   Compiling hex v0.3.2
   Compiling rustc-workspace-hack v1.0.0
   Compiling shell-escape v0.1.4
   Compiling thread_local v0.3.6
   Compiling rand_core v0.3.1
   Compiling rand_jitter v0.1.4
   Compiling humantime v1.2.0
   Compiling rand_hc v0.1.0
   Compiling rand_xorshift v0.1.1
   Compiling rand_isaac v0.1.1
   Compiling num-traits v0.2.8
   Compiling num-integer v0.1.41
   Compiling rand_chacha v0.1.1
   Compiling rand_pcg v0.1.2
   Compiling rand v0.6.5
   Compiling aho-corasick v0.7.6
   Compiling quote v0.6.13
   Compiling quote v1.0.2
   Compiling time v0.1.42
   Compiling dirs-sys v0.3.4
   Compiling atty v0.2.13
   Compiling rand_os v0.1.3
   Compiling directories v2.0.2
   Compiling regex v1.2.1
   Compiling chrono v0.4.7
   Compiling backtrace-sys v0.1.31
   Compiling env_logger v0.6.2
   Compiling vergen v3.0.4
   Compiling synstructure v0.10.2
   Compiling backtrace v0.3.35
   Compiling serde_derive v1.0.99
   Compiling failure v0.1.5
   Compiling miri v0.1.0 (/home/wesley/code/rust/miri)
   Compiling semver v0.9.0
   Compiling serde_json v1.0.40
   Compiling rustc_version v0.2.3
   Compiling cargo_metadata v0.8.2
    Finished release [optimized] target(s) in 1m 14s
    Finished release [optimized] target(s) in 0.02s
     Running `target/release/cargo-miri miri setup`
    Updating crates.io index
   Compiling compiler_builtins v0.1.19
   Compiling core v0.0.0 (/home/wesley/code/rust/rust/src/libcore)
   Compiling cc v1.0.41
   Compiling libc v0.2.62
   Compiling build_helper v0.1.0 (/home/wesley/code/rust/rust/src/build_helper)
   Compiling cmake v0.1.42
   Compiling unwind v0.0.0 (/home/wesley/code/rust/rust/src/libunwind)
   Compiling rustc_tsan v0.0.0 (/home/wesley/code/rust/rust/src/librustc_tsan)
   Compiling rustc_asan v0.0.0 (/home/wesley/code/rust/rust/src/librustc_asan)
   Compiling rustc_lsan v0.0.0 (/home/wesley/code/rust/rust/src/librustc_lsan)
   Compiling rustc_msan v0.0.0 (/home/wesley/code/rust/rust/src/librustc_msan)
   Compiling std v0.0.0 (/home/wesley/code/rust/rust/src/libstd)
   Compiling rustc-std-workspace-core v1.0.0 (/home/wesley/code/rust/rust/src/tools/rustc-std-workspace-core)
   Compiling cfg-if v0.1.8
   Compiling alloc v0.0.0 (/home/wesley/code/rust/rust/src/liballoc)
   Compiling panic_abort v0.0.0 (/home/wesley/code/rust/rust/src/libpanic_abort)
   Compiling rustc-std-workspace-alloc v1.0.0 (/home/wesley/code/rust/rust/src/tools/rustc-std-workspace-alloc)
   Compiling panic_unwind v0.0.0 (/home/wesley/code/rust/rust/src/libpanic_unwind)
   Compiling hashbrown v0.5.0
    Finished release [optimized] target(s) in 28.99s
   Compiling unicode-width v0.1.6
   Compiling term v0.0.0 (/home/wesley/code/rust/rust/src/libterm)
   Compiling proc_macro v0.0.0 (/home/wesley/code/rust/rust/src/libproc_macro)
error[E0463]: can't find crate for `core`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `unicode-width`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `proc_macro`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `term`.

To learn more, run the command again with --verbose.
error: `"/home/wesley/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/cargo" "build" "--release" "--manifest-path" "/tmp/xargo.kHu0e3jhZph2/Cargo.toml" "--target" "x86_64-unknown-linux-gnu" "-p" "test"` failed with exit code: Some(101)
note: run with `RUST_BACKTRACE=1` for a backtrace
fatal error: Failed to run xargo
wesley@endurance:~/code/rust/miri>

RalfJung · 2019-08-29T12:29:09Z

That looks like your xargo is outdated. What does xargo --version say? Are you sure you are using the latest master of Miri?

wesleywiser · 2019-09-01T15:18:56Z

Are you using the latest master of miri?

Yes, I just rebased my changes on master (3a5c44c2d83b04651a2e04178087e23e3db6fccc).

That looks like your xargo is outdated.

I don't remember ever installing xargo so I assume ./miri did that?

> xargo --version
xargo 0.3.16
cargo 1.38.0-nightly (677a180f4 2019-07-08)

wesleywiser · 2019-09-01T15:38:31Z

This is very strange. If I use the latest nightly (2019-08-31), ./miri test works fine. Furthermore, if I re-enable the hashmap tests, they also pass under the latest nightly. So I guess this is fixed?

~~I'll submit a PR to re-enable the tests and verify it works on the CI.~~

Edit: I just realized the tests are passing on macOS on CI and I'm currently using my MacBook. I'll retest this on Linux.

Edit 2: I am seeing the test failure on Linux. I think perhaps I'm missing some kind of configuration necessary to use miri with a local rustc build. If I use a nightly, then miri compiles fine but if I use my local copy (with up to date HEAD and xargo), then I get the above errors.

RalfJung · 2019-09-01T19:55:51Z

./miri test does not use the bad opt-level any more. I didn't want to hold up Miri on this issue so I reduced the level we are using for testing.

Try this:

RUST_BACKTRACE=1 ./miri run-debug tests/run-pass/hashmap.rs -Zmir-opt-level=2

wesleywiser · 2019-09-01T20:09:52Z

Thanks, that repro's on nightly.

I tried running various parts of the ./miri script by hand and I noticed that if I run cargo run --release --bin cargo-miri -- miri setup on my stage2 toolchain, I get this:

    Finished release [optimized] target(s) in 0.07s
     Running `target/release/cargo-miri miri setup`
Installing rust-src component: `rustup component add rust-src`
error: toolchain 'stage2' does not support components
info: caused by: invalid toolchain name: 'stage2'
fatal error: Failed to install rust-src component

Should it be trying to install the rust-src component? Do I need to do something special to get that component to point to my local repo?

RalfJung · 2019-09-01T20:19:37Z

You need to set XARGO_RUST_SRC to the src dir of your Rust toolchain. Components don't work with linked toolchains.

Also, the HashMap test indeed takes a somewhat different code path on macOS due to #686, which might explain the platform differences.

wesleywiser · 2019-09-01T20:39:47Z

Hmm... the error looks the same as before even with XARGO_RUST_SRC set:

    Finished release [optimized] target(s) in 28.55s
   Compiling unicode-width v0.1.6
   Compiling proc_macro v0.0.0 (/home/wesley/code/rust/rust/src/libproc_macro)
   Compiling term v0.0.0 (/home/wesley/code/rust/rust/src/libterm)
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `term`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `std`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `proc_macro`.
warning: build failed, waiting for other jobs to finish...
error[E0463]: can't find crate for `core`

error: aborting due to previous error

For more information about this error, try `rustc --explain E0463`.
error: Could not compile `unicode-width`.

To learn more, run the command again with --verbose.
error: `"/home/wesley/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/cargo" "build" "--release" "--manifest-path" "/tmp/xargo.mEQSIPsXjGY4/Cargo.toml" "--target" "x86_64-unknown-linux-gnu" "-p" "test"` failed with exit code: Some(101)
note: run with `RUST_BACKTRACE=1` for a backtrace
fatal error: Failed to run xargo

RalfJung · 2019-09-01T20:55:45Z

What exactly did you set XARGO_RUST_SRC to? It should end in src -- it must be the dir containing libstd etc

wesleywiser · 2019-09-01T21:04:15Z

XARGO_RUST_SRC=~/code/rust/rust/src

wesley@endurance:~/code/rust/miri> ls ~/code/rust/rust/src
bootstrap     liballoc        libpanic_unwind       librustc_ast_borrowck     librustc_errors       librustc_llvm      librustc_passes         librustc_traits  libsyntax      llvm-emscripten  stage0.txt
build_helper  libarena        libproc_macro         librustc_codegen_llvm     librustc_fs_util      librustc_lsan      librustc_plugin         librustc_tsan    libsyntax_ext  llvm-project     stdarch
ci            libcore         libprofiler_builtins  librustc_codegen_ssa      librustc_incremental  librustc_macros    librustc_privacy        librustc_typeck  libsyntax_pos  README.md        test
doc           libfmt_macros   librustc              librustc_codegen_utils    librustc_interface    librustc_metadata  librustc_resolve        librustdoc       libterm        rtstartup        tools
etc           libgraphviz     librustc_apfloat      librustc_data_structures  librustc_lexer        librustc_mir       librustc_save_analysis  libserialize     libtest        rustc
grammar       libpanic_abort  librustc_asan         librustc_driver           librustc_lint         librustc_msan      librustc_target         libstd           libunwind      rustllvm

RalfJung · 2019-09-02T10:20:11Z

Very strange. Depending in your shell, I think `~` might not work when setting env vars. Try giving a fully explicit absolute path?

wesleywiser · 2019-09-02T13:46:48Z

Very strange.
Depending in your shell, I think ~ might not work when setting env vars. Try giving a fully explicit absolute path?

It doesn't seem to make a difference.

Yeah, I'm kind of stumped. I've tried it on two different computers and two different OS's and I can't seem to get it to work on either. I don't think I have anything especially strange installed or configured on either.

RalfJung · 2019-09-02T14:11:09Z

If I use a nightly, then miri compiles fine

And the test suite also passes?

cargo 1.38.0-nightly (677a180f4 2019-07-08)

This looks like a fairly old nightly version of Rust.
But I guess you are using the latest rustc and Miri master for testing the self-compiled builds?

if I use my local copy (with up to date HEAD and xargo), then I get the above errors.

Miri HEAD or rustc HEAD? What's your rustc HEAD? Did you try using the latest rustc master for that (so that it is a version I can also try and reproduce this with)?

For the record, here's the not-commented-out part of my config.toml for rustc:

[llvm]
assertions = true
ninja = true
[build]
submodules = false
[rust]
debug-assertions = true
debuginfo-level = 1
incremental = true
deny-warnings = false

oli-obk · 2019-09-02T14:45:00Z

Note that you need to turn optimized run pass tests back on in order to see the failure. I just tried it and it still fails on nightly

wesleywiser · 2019-09-02T14:46:18Z

And the test suite also passes?

With a clean checkout of miri? Yes. If I re-enable the MIR optimizations then the hashmap test fails.

This looks like a fairly old nightly version of Rust.
But I guess you are using the latest rustc and Miri master for testing the self-compiled builds?

I have a checkout of rustc from just after rust-lang/rust#63580 was merged. I'm going to pull and rebase my change right now. I have a build started on top of fdaf594bab31eec75fb6d582cd33e5a5b43de7f4 which is the current HEAD on master of rust-lang/rust. My checkout of miri is at 3a5c44c2d83b04651a2e04178087e23e3db6fccc which is the current master.

RalfJung · 2019-09-02T14:47:11Z

The issue here is @wesleywiser can't even run the test suite as xargo fails.

I have a checkout of rustc from just after rust-lang/rust#63580 was merged.

Ah, that might be it. We had to update Miri when rust-lang/rust#63637 landed and that means rustc prior to that don't work any more.

wesleywiser · 2019-09-02T15:01:01Z

In ~/code/rust/rust I have fdaf594bab31eec75fb6d582cd33e5a5b43de7f4 checked out with a change I'm testing on top of that.

I ran ./x.py build

That completed successfully.

In ~/code/rust/miri I have 3a5c44c2d83b04651a2e04178087e23e3db6fccc checked out with a change to run the tests with optimization on top of that.

I ran rustup override set stage2.
I ran cargo clean.
I ran XARGO_RUST_SRC=/home/wesley/code/rust/rust/src ./miri test

That worked!

wesleywiser · 2019-09-02T15:02:03Z

So the issue was almost certainly my outdated rustc checkout. Sorry for wasting your time!

RalfJung · 2019-09-02T15:09:56Z

My rustc test build to reproduce is just done. I guess I don't need that any more then. ;)

Good to hear it works now. :)

XARGO_RUST_SRC=/home/wesley/code/rust/rust/src ./miri test

In theory, even just ./miri test should work. Could you try that? (Do rm ~/.cache/miri/ -rf to make sure it actually builds the thing and does not use a cached version.)

wesleywiser · 2019-09-02T19:46:16Z

In theory, even just ./miri test should work. Could you try that? (Do rm ~/.cache/miri/ -rf to make sure it actually builds the thing and does not use a cached version.)

Yep, that works great!

Perhaps I should add an entry to the "Common Problems" section of the Readme with this info?

Also, I just opened rust-lang/rust#64100 which will fix the regression. I believe I need to wait for that PR to land and then open a PR here with a rustup right?

RalfJung · 2019-09-02T21:04:03Z

Perhaps I should add an entry to the "Common Problems" section of the Readme with this info?

This concrete issue is unlikely to still come up in a month or two because old compilers will get used less and less.
Do you have a good idea about where to put the general advice of "use the latest rustc" (or the one in rust-version)? We already say something along those lines but obviously not strongly enough. ;)

wesleywiser · 2019-09-02T21:42:26Z

I was going to add something like:

error[E0463]: can't find crate for `std`

This may indicate that your local checkout of rust is out of date. Try to update both rust and miri to the latest version with git pull.

RalfJung · 2019-09-02T21:44:28Z

This is unlikely to be the error you get with other Miri/rustc version mismatches. And IIRC you also get that error when using a bogus XARGO_RUST_SRC. So I don't think that specifically would still be useful advice in, say, 2 months.

wesleywiser · 2019-09-02T21:51:43Z

Ah, ok. Feel free to disregard then. :) Given how much back and forth it took to figure that out, I was hoping a warning might prevent this kind of thing in the future.

RalfJung · 2019-12-29T11:39:39Z

Final bullet fixed by #1133

RalfJung mentioned this issue Oct 15, 2018

run tests with optimized MIR #481

Closed

RalfJung added the C-bug Category: This is a bug. label Nov 17, 2018

RalfJung added the A-tests Area: affects our test suite or CI label Mar 8, 2019

RalfJung changed the title ~~Enable test suite for MIR optimizations again~~ Enable MIR optimizations in test suite again Nov 4, 2019

RalfJung closed this as completed Dec 29, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Enable MIR optimizations in test suite again #480

Enable MIR optimizations in test suite again #480

RalfJung commented Oct 15, 2018 •

edited

Loading

RalfJung commented Oct 31, 2018

RalfJung commented Aug 27, 2019

wesleywiser commented Aug 27, 2019

wesleywiser commented Aug 29, 2019

RalfJung commented Aug 29, 2019 •

edited

Loading

wesleywiser commented Aug 29, 2019

RalfJung commented Aug 29, 2019

wesleywiser commented Sep 1, 2019

wesleywiser commented Sep 1, 2019 •

edited

Loading

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 2, 2019 via email

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019 •

edited

Loading

oli-obk commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019 •

edited

Loading

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Dec 29, 2019

Enable MIR optimizations in test suite again #480

Enable MIR optimizations in test suite again #480

Comments

RalfJung commented Oct 15, 2018 • edited Loading

RalfJung commented Oct 31, 2018

RalfJung commented Aug 27, 2019

wesleywiser commented Aug 27, 2019

wesleywiser commented Aug 29, 2019

RalfJung commented Aug 29, 2019 • edited Loading

wesleywiser commented Aug 29, 2019

RalfJung commented Aug 29, 2019

wesleywiser commented Sep 1, 2019

wesleywiser commented Sep 1, 2019 • edited Loading

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 1, 2019

wesleywiser commented Sep 1, 2019

RalfJung commented Sep 2, 2019 via email

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019 • edited Loading

oli-obk commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019 • edited Loading

wesleywiser commented Sep 2, 2019

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

error[E0463]: can't find crate for `std`

RalfJung commented Sep 2, 2019

wesleywiser commented Sep 2, 2019

RalfJung commented Dec 29, 2019

RalfJung commented Oct 15, 2018 •

edited

Loading

RalfJung commented Aug 29, 2019 •

edited

Loading

wesleywiser commented Sep 1, 2019 •

edited

Loading

RalfJung commented Sep 2, 2019 •

edited

Loading

RalfJung commented Sep 2, 2019 •

edited

Loading