On 30/05/17 01:03, Junio C Hamano wrote:
> Ramsay Jones <ramsay@xxxxxxxxxxxxxxxxxxxx> writes:
>
>> See commit c7018be509 ("test: allow skipping the remainder", 18-05-2017)
>> which is currently merged to the 'next' branch (merge 03b8a61e47 of the
>> 'jc/skip-test-in-the-middle' branch).
>>
>> (see also http://testanything.org)
>>
>> If you look at http://testanything.org//tap-specification.html, it shows
>> that you are allowed to annotate a plan of '1..0' with a SKIP directive
>> to explain why no tests in a file were run. However, a plan with '1..n'
>> (for any n > 0) must not have any annotation. (Back in 2012, when I wrote
>> commit bf4b721932, I found much better documentation than the above!)
>>
>> So, after commit c7018be509, you can now use the 'skip_all' facility
>> after having run some tests; it now converts that into an 'SKIP comment'
>> just before the test plan, effectively skipping the remainder of the
>> tests in the file. (since we are using a 'trailing plan', and have thus
>> not declared how many tests we had intended to run, we can output an
>> accurate plan).
>
> Yes, but I consider that c7018be509 is an ugly workaround, not a
> part of a properly designed framework. Unless it is absolutely
> necessary to run some tests before we may conditionally want to say
> "skip_all/test_done", we should strive to add tests _after_ these
> conditional skil_all/test_done is done.
yes, I don't disagree with that (which is why I said that I would
have split t5545 into two files). ;-)
> In this case, I do not see there is a strong reason why the new test
> must come before the "setup" test. Sure, it does not use UNCPATH so
> the new test may be able to run even when the current path cannot be
> spelled as UNC, but that is a natural fallout of (ab)using the test
> script that was meant for UNC testing for something else, so I think
> a proper way out would be either (1) to use a separate script, if
> the new test wants to run whether UNC path can be determined,
Yes, I had intended to suggest this (or an existing script, protected
by the MINGW prerequisite), but forgot!
> or (2)
> just accept the fact that the new test will only be run when UNC
> paths are tested.
I prefer (1).
> Relying on the hack c7018be509 did is much less
> appealing.
ATB,
Ramsay Jones