raco test: set current-test-invocation-directory

[benknoble]

Running "raco test foo/bar/test.rkt" will cause Racket to change directory to "foo/bar" before invoking "test.rkt", and error messages are printed relative to (current-directory) like

raco test: "foo/bar/test.rkt"
--------------------
fail
FAILURE
name:       check-equal?
location:   test.rkt:5:0
actual:     1
expected:   2
--------------------
1/1 test failures

For an editor to consume this, it has to post-process the output to correlate "raco test: path" with "location: partial path" (which is difficult when running tests in parallel). This is not ideal; output with

location:   foo/bar/test.rkt:5:0

is immediately useable by editors for jumping to the error.

Expose the original directory in which "raco test" was launched so that rackunit (or other test frameworks) can appropriately adjust filepaths when reporting errors.

Checklist

• Bugfix
• Feature
• tests included
• documentation

[benknoble]

I believe the new version addresses @usaoc's comments. Should the out contract use path-for-some-system? since that's what the guard actually produces, or is path? sufficient documentation here?

range-diff:

1:  216c2bd1f5 ! 1:  df01b0d1dc raco test: set current-test-invocation-directory
    @@ Commit message
         that rackunit (or other test frameworks) can appropriately adjust
         filepaths when reporting errors.
     
    +    Also add a name to the test-log-enabled? parameter while we're here.
    +
      ## pkgs/compiler-lib/compiler/commands/test.rkt ##
     @@ pkgs/compiler-lib/compiler/commands/test.rkt: (define-values (sum _ deps)
        "Save stdout and stderr to file, overwrite if it exists."
    @@ pkgs/compiler-lib/raco/testing.rkt
     +;; records the original "raco test" directory while raco test changes directory
     +;; to invoke tests
     +(define current-test-invocation-directory
    -+  (make-parameter #f))
    ++  (make-parameter
    ++   #f
    ++   (λ (path)
    ++     (cond
    ++       [(path-string? path) (path->directory-path (simplify-path path))]
    ++       [(not path) path]
    ++       [else (raise-argument-error 'current-test-invocation-directory
    ++                                   "(or/c #f path-string?)"
    ++                                   path)]))
    ++   'current-test-invocation-directory))
      
      (define test-log-enabled?
    -   (make-parameter #t (lambda (v) (and v #t))))
    +-  (make-parameter #t (lambda (v) (and v #t))))
    ++  (make-parameter #t (lambda (v) (and v #t)) 'test-log-enabled?))
    + 
    + (define TOTAL 0)
    + (define FAILED 0)
     
      ## pkgs/racket-doc/scribblings/raco/test.scrbl ##
     @@ pkgs/racket-doc/scribblings/raco/test.scrbl: with @exec{raco test} should also use this library to log test results.
       counted in the test log, such as when testing a custom check's failure
       behavior.}
      
    -+@defparam[current-test-invocation-directory test-invocation-directory (or/c #f path-string?) #:value #f]{
    ++@defparam*[current-test-invocation-directory
    ++            path
    ++            (or/c #f path-string?)
    ++            (or/c #f path?)
    ++            #:value #f]{
     +Contains the directory from which tests were invoked by, @emph{e.g.}, @exec{raco
     +test}. This may differ from @racket[current-directory] when the test runner
     +changes directory before invoking a specific test file and should be set by test

[usaoc]

I think a path or string will always end up as a path for the current platform, unless you explicitly produce a path for a different platform, so path? is correct here.

[rfindler]

Sorry if I am missing something, but on the input side would path-string? Be acceptable?

…

On Sat, May 11, 2024 at 8:43 AM Wing Hei Chan ***@***.***> wrote: I think a path or string will always end up as a path for the current platform, unless you explicitly produce a path for a different platform, so path? is correct here. — Reply to this email directly, view it on GitHub <#4988 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AADBNMHH3WXZN5ECONSL7QDZBYOAHAVCNFSM6AAAAABHLTKXMSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMBVG43DGMBQGY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[benknoble]

Sorry if I am missing something, but on the input side would path-string? be acceptable?

Yes; indeed, the documentation and code (should) both agree that the parameter accepts (or/c #f path-string?). The output coerces to (or/c #f path?).

[benknoble]

@mflatt it looks like I've missed a case somewhere: when I do raco test -j 8 foo/bar/test.rkt, I still get the new output. But with raco test -j 8 foo/bar/test{1,2}.rkt, I get the old output (meaning the new parameter hasn't been set, I think).

Any pointers?

[mflatt]

@benknoble With -j 8, parallelism can be implemented with a thread, place, or process. For places, maybe line 121 of test.rkt is the right point to set the parameter in the new place, and maybe the value to be used for the parameter needs to be passed to that place. For processes, probably the value of the parameter needs to be passed as a command-line argument to be received by the process submodule.

[benknoble]

Thanks. I started thinking along those lines; the code references are helpful.