Protocol method placeholders marked as untested #1616

I see what you mean that these lines can't be executed. But I'm not sure how coverage.py could know that? You can exclude lines from coverage measurement with the exclude_also configuration setting.

@nedbat perhaps a recommended pre commit hook that adds those functions / lines would be of help (or a helper to dummy invoke them in testsuites)

There are other issues about coverage.py not understanding various typing constructs:

• Assume if TYPE_CHECKING: ... else: ... block is covered #831
• Coverage on overload stubs for typing #970

So perhaps it is time to add some typing patterns to the default exclude patterns.

As an experiment, does this work to exclude the Protocols in your project?

[report]
exclude_also = 
    class \w+\(Protocol\):

Note that protocol abc classes ought not to be excluded as they sometimes implement default implementations of

(your comment is a bit truncated, but:) Can you show a small sample of code that shouldn't be excluded? This is the risk of a simplistic method like regexes for deciding things about code.

I believe a fair set would be methods whose body is only the ellipsis

I believe a fair set would be methods whose body is only the ellipsis

Unfortunately, I think that's not possible to express with a line-based regex. Is there a way to extend the power without becoming a full-fledged AST pattern matcher?

BTW, links to real-world examples of tricky Protocols would be helpful.

@nedbat i believe that a reasonable hackish approximation would be to just consider lines with only a ellipsis covered

It's slightly unfortunate to me that black or whatever other linters want the ... treated like a normal function body (and put on their own line), otherwise I think even "better" would be to limit this to :.*\.\.\.$ and to encourage method bodies to not have their own line for Protocols.

But given they do indeed put the ellipsis on their own line I guess I too agree it's probably most reasonable at the minute to do ^^.

@nedbat would it be sensible to ignore all single lines that contain only an ellipsis? i vaugely recall that function definitions get covered at definition time, so the ellipsis should be the only missing covered item

Color me ... well not shocked but certainly pleasantly surprised that

It's slightly unfortunate to me that black or whatever other linters want the ... treated like a normal function body

is now "fixed" / "improved" in Black 24 via psf/black#1797.

I'm not sure if this was closed because @nedbat WONTFIXed it (obviously all good if so) or if @andrewlalis you just closed it yourself because it was remaining open. But perhaps it's worth another consideration in light of the formatter change.