Go: cmd/go: add go test -skip to skip specific tests

Created on 23 Sep 2020  Â·  26Comments  Â·  Source: golang/go

t.Skip() is an option but it requires changing the test code.

It's hard to write a correct regex to pass to the -run flag in order to skip specific tests. This is due to the lack of lookaheads in the regexp library. Previous discussion here: https://groups.google.com/g/golang-nuts/c/7qgSDWPIh_E?pli=1

Proposal Proposal-Accepted Proposal-FinalCommentPeriod WaitingForInfo

Most helpful comment

Do you have a specific suggestion?

I didn't want to prescribe a solution but here's some off the top of my head

1) expand run flag functionality - -run [regex] to allow look aheads - thus negation

  • pros

    • most people are familiar with this flag

  • cons

    • requires expanded regex implementation

2) new flag -skip [regex]

  • pros

    • mirrors run flag

    • may be easier to use than a negative regex

  • cons

    • requires definition of precedence between -skip& -run

    • it'd be nice to see skipped test output - that might require more work to implement

To run specific tests you can already write go test -run="TestOne|TestTwo|TestThree".

This doesn't scale if I have O(10s) of tests in a package and want to skip one or two

All 26 comments

Do you have a specific suggestion?

To run specific tests you can already write go test -run="TestOne|TestTwo|TestThree".

Do you have a specific suggestion?

I didn't want to prescribe a solution but here's some off the top of my head

1) expand run flag functionality - -run [regex] to allow look aheads - thus negation

  • pros

    • most people are familiar with this flag

  • cons

    • requires expanded regex implementation

2) new flag -skip [regex]

  • pros

    • mirrors run flag

    • may be easier to use than a negative regex

  • cons

    • requires definition of precedence between -skip& -run

    • it'd be nice to see skipped test output - that might require more work to implement

To run specific tests you can already write go test -run="TestOne|TestTwo|TestThree".

This doesn't scale if I have O(10s) of tests in a package and want to skip one or two

hey @ianlancetaylor - do you have any additional questions or want me to clarify anything above?

No, I'm good.

This issue is in the proposal process now and will get into the proposal committee in due course. The committee looks at a lot of issues (see minutes at #33502), so it won't be immediately, but it shouldn't take too long.

Thanks.

Almost no one on earth understands how to write negation regexps correctly, so I would lean toward -skip if we do this.

requires definition of precedence between -skip & -run

I can see only two reasonable definitions.

  1. Define that the default value of -run matches everything, and therefore -skip takes precedence over -run.
  2. Define that the default value of -run imposes no constraints, and therefore it is an error to specify values of -run and -skip that both match the same test.

I have a slight preference for (1), because it treats the default value of the -run flag the same as other values.

Does precedence mean the order the filters are applied in or which is used if both are given?

If it's the order, then -run followed by -skip makes the most sense to me and seems more useful than disallowing the combination.

Yeah, it seems especially useful in the context of subtests:

go test foo -run=TestBar -skip=TestBar/flaky_subtest

Does precedence mean the order the filters are applied in or which is used if both are given?

The latter in my opinion

I have almost moved some subtests into their own top level test function on a few occasions exactly because there was not a good way to skip some subset of them. I have also had to run go test -list iterations just to figure out a regex that would run all tests except a select few, and the resulting regex is often overly long given the use case of wanting to skip a small subset when a package has a lot of tests.

It sounds like people are generally in favor of adding -skip with a default of "matches nothing at all".
The -skip setting applies after the -run setting (which already defaults to "match everything")
and only applies to tests (not benchmarks).

Do I have that right? Does anyone object to adding -skip as described?

The -skip setting … only applies to tests (not benchmarks).

Hmm. It would be unfortunate to have an easy way to skip specific tests but not specific benchmarks.

I think it would make sense to either have -skip apply to both tests and benchmarks, or to add a separate -benchskip or similar flag for those. (It seems a bit simpler to me to make -skip also apply to benchmarks, but I don't have a strong preference either way.)

Do I have that right? Does anyone object to adding -skip as described?

Sounds good to me

Hmm. It would be unfortunate to have an easy way to skip specific tests but not specific benchmarks.

Expanding the scope of -skip to cover benchmarks makes sense to me. I don't mind if this comes later if it's not a trivial thing to do.

A way to skip benchmarks would be welcome as well.

The -run and -bench flags have different defaults. In my experience go test -run=^$ -bench=. is pretty common today, but would people start trying to use go test -skip=. -bench=. to mean the same thing? It seems to me a separate -benchskip flag would avoid cross pollinating the two sets and reduce the chances for surprise.

Instead of introducing a new flag, could we add some syntax to the existing flag? Let a pattern starting with ! invert the sense of the match. So go test -run=!Foo|Bar would run all tests that don't match Foo or Bar, same as the proposed -skip flag.

This avoids introducing new flags (-skip and -benchskip) and avoids issues about whether -run or -skip takes precedence. -run already has special handling for / so it is already not a straight regexp.

-run already has special handling for / so it is already not a straight regexp.

Truthfully I was tripped up by this when I first encountered it

@magical, further complications in the syntax for the -skip flag would be pretty awkward for the “run a specific test but skip a specific subtest” use-case mentioned previously.

@ChrisHines, I think the question is, would someone running go test -bench=BenchmarkFoo -skip=BenchmarkFoo/some_subbench expect it to skip BenchmarkFoo/some_subbench?

I would argue that pretty much everyone would expect that to skip the sub-benchmark, so I suspect that a unified -skip flag would be less confusing than a separate -benchskip flag.

The -run and -bench flags have different defaults. In my experience go test -run=^$ -bench=. is pretty common today

FWIW this is usually a bad idea, at least interactively, since you inevitably end up benchmarking broken code. (And boy can you make broken code run fast!)

@magical, more magic in -run is not a good idea. The slash was not great either but is done.

The -run and -bench flags have different defaults. In my experience go test -run=^$ -bench=. is pretty common today

FWIW this is usually a bad idea, at least interactively, since you inevitably end up benchmarking broken code. (And boy can you make broken code run fast!)

That's a fair point. I guess it depends on one's workflow. In my experience testing and benchmarking are distinct steps of the workflow. If the tests fail I don't want to wait for benchmarks to run since they would be meaningless. Once the tests pass and I am benchmarking I don't want to waste time rerunning the tests before getting the benchmark data.

-list already applies to everything - tests, benchmarks, examples - so I think it's probably OK to have -skip do the same (apply to benchmarks; tests and examples were already covered since they are controlled by -run).

Does anyone object to doing this?

No objection here.

No objection

On Wed, Oct 21, 2020 at 14:36 Chris Hines notifications@github.com wrote:

No objection here.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/golang/go/issues/41583#issuecomment-713785061, or
unsubscribe
https://github.com/notifications/unsubscribe-auth/AAAERAQIEIIMXURBJQH2EMDSL4S3XANCNFSM4RXF7DKQ
.

Based on the discussion, this seems like a likely accept.

No change in consensus, so accepted.

Was this page helpful?
0 / 5 - 0 ratings