Add bless x.py subcommand for easy ui test replacement

[oli-obk]

fixes#49815

r? @nikomatsakis

[oli-obk]

cc @ehuss

[kennytm]

With the implementation in this PR, I think it should be a --bless flag to test instead of an entire new subcommand.

$ ./x.py test src/test/ui --bless

In #49815's suggestion the subcommand works like update-reference.sh. These scripts won't run tests themselves, they only copy the generated stdout/stderr files, meaning the workflow would be like

$ ./x.py test src/test/ui(... fix errors ...)
$ ./x.py test src/test/ui(... fix errors ...)
$ ./x.py test src/test/ui(done)
$ ./x.py bless src/test/ui # <- will not run tests again

[petrochenkov]

In my experience the workflow is

// 1. Initial run
x.py test ui
// 2. Update stderr/stdout
update-all-reference.sh
// 3. Was it enough or some `//~ ERROR` annotations need update as well?
x.py test ui
// 4. Optionally: Fix `//~ ERROR` annotations
(... fix errors ...)
// 5. Was it enough or stderr/stdout need updating again?
x.py test ui
// 6. Optionally: Update stderr/stdout
update-all-reference.sh
// 7. Make sure everything passes now, maybe go back to `4.` if not
x.py test ui

so I'd very much prefer to do the "test-update-test" operation in one command to keeping the "update" logic separate.
Then the workflow would turn into

// 1. Initial "test-update-test" run
x.py bless ui
// 2. Optionally: Fix `//~ ERROR` annotations
(... fix errors ...)
// 3. Second "test-update-test" run to make sure everything passes now, maybe go back to `2.` if not
x.py bless ui

I'm not sure about test --bless vs bless, the first one is more correct because it's "test with a tweak", but the second one is more convenient.

[oli-obk]

Yes that is also my workflow. I changed it to a flag, as that makes the code simpler and I already pass such a mess of flags to x.py it doesn't make a difference to me.

RUST_BACKTRACE=1 ./x.py test --stage 1 src/test/ui --test-args const_signed_pat --bless

[rust-highfive]

The job x86_64-gnu-llvm-3.9 of your PR failed on Travis (raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

[00:49:31] Compiling ansi_term v0.11.0
[00:49:31] Compiling difference v2.0.0
[00:49:32] Compiling pretty_assertions v0.5.1
[00:49:32] Compiling bootstrap v0.0.0 (file:///checkout/src/bootstrap)
[00:49:37] error[E0063]: missing field `bless` in initializer of `flags::Subcommand`
[00:49:37] --> bootstrap/builder.rs:1450:22
[00:49:37] |
[00:49:37] 1450 | config.cmd = Subcommand::Test {
[00:49:37] | ^^^^^^^^^^^^^^^^ missing `bless`
[00:49:37] error: aborting due to previous error
[00:49:37] [00:49:37] For more information about this error, try `rustc --explain E0063`.
[00:49:37] For more information about this error, try `rustc --explain E0063`.
[00:49:37] error: Could not compile `bootstrap`.
[00:49:37] To learn more, run the command again with --verbose.
[00:49:37] [00:49:37] [00:49:37] command did not execute successfully: "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/cargo" "test" "--"
[00:49:37] command did not execute successfully: "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/cargo" "test" "--"
[00:49:37] expected success, got: exit code: 101
[00:49:37] [00:49:37] [00:49:37] failed to run: /checkout/obj/build/bootstrap/debug/bootstrap test
[00:49:37] Build completed unsuccessfully in 0:00:21
[00:49:37] make: *** [check] Error 1
[00:49:37] Makefile:58: recipe for target 'check' failed
The command "stamp sh -x -c "$RUN_SCRIPT"" exited with 2.
travis_time:start:045cdfe4
$ date && (curl -fs --head https://google.com | grep ^Date: | sed 's/Date: //g' || true)

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

[ehuss]

Sweet! Sorry I didn't get to this sooner! (BTW, love the branch name.)

Maybe the error message that says "To update references, run this command..." should be updated to the new command?

Also, beware that there are a bunch of path changes coming in #50400. There shouldn't be any fundamental problems, but some of the function names will change.

[ehuss]

This doesn't seem to work for nll files. For an example, copy E0508.rs to src/test/ui and try to bless it. You should end up with these three files:

E0508.ast.nll.stderr
E0508.ast.stderr
E0508.mir.stderr

I think the issue is that expected_output_path doesn't include mode if the file doesn't already exist. I think the solution is to call the global expected_output_path (not the self version).

[nikomatsakis]

@ehuss -- you're the one most familiar with the code at this point, would you be up for tackling the review here? (I can delegate to you from bors)

[ehuss]

@nikomatsakis I don't mind, but I don't have any more comments beyond what I posted.

[oli-obk]

@bors delegate=ehuss

[bors]

✌️ @ehuss can now approve this pull request

[killercup]

@oli-obk want to add replacement of .fixed files, too? (not necessarily in this pr though)

[bors]

☔ The latest upstream changes (presumably #50400) made this pull request unmergeable. Please resolve the merge conflicts.

[oli-obk]

@killercup that should automagically work with this PR ;) There's no code specific to stderr files in it.

[nikomatsakis]

@oli-obk are you ready for a re-review from @ehuss then?

[oli-obk]

Yea, I was just waiting for travis :) Last time I broke some tests that ran very late in the pipeline.

[ehuss]

@bors: r+

Nice!

[bors]

📌 Commit 0ac2fd1 has been approved by ehuss

[Mark-Simulacrum]

@bors rollup

[scottmcm]

Thank you especially from those of us who run on windows without a shell to run the .sh 🎉