Fabien COELHO <coelho@cri.ensmp.fr> writes:
>>> I run the test, figure out the number it found in the resulting
>>> error message, and update the number in the source.
>> Yeah, but then what error is all that work protecting you from?
> I'm not sure I understand your point. I agree that Perl doing the counting
> may hide issues. Now it is more of an incremental thing, if a test is
> added the counter is upgraded accordingly, and the local consistency can
> be checked.
Well, IIUC the point of "tests => nnn" is to protect against the
possibility that some coding mistake caused some of the tests to not
get run. If you change the test script and then take Perl's word for
what nnn should be, what protection have you got that you didn't introduce
such a mistake along with whatever your intended change was? It seems
pointless to maintain the tests count that way.
> Anyway, as some tests may have to be skipped on some platforms, it seems
> that the done_testing approach is sounder. The v12 patch uses that.
Sounds good.
regards, tom lane