Re: [GSoC] microporject test_path_is_*

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Mar 27, 2019, 14:10 UTC
Message-ID: <8736n8be71.fsf@evledraar.gmail.com>
In-Reply-To: <20190327121411.GE32732@szeder.dev>

On Wed, Mar 27 2019, SZEDER Gábor wrote:

Show 38 quoted lines

> On Wed, Mar 27, 2019 at 12:21:55PM +0100, Ævar Arnfjörð Bjarmason wrote:
>>
>> On Wed, Mar 27 2019, SZEDER Gábor wrote:
>>
>> > On Wed, Mar 27, 2019 at 11:09:18AM +0100, Ævar Arnfjörð Bjarmason wrote:
>> >> > There are likewise several that use one of
>> >> >    ! test -e path/to/filename
>> >> > or
>> >> >    ! test -f path/to/filename
>> >> > or
>> >> >   test ! -f path/to/filename
>> >> > which could be replaced by
>> >> >   test_path_is_missing path/to/filename
>> >>
>> >> Interesting that for some we use the 'test_is_there/test_is_not_there'
>> >> pattern and for others 'test_is_there [!]'. E.g
>> >> test_path_exist/test_path_is_missing v.s. test_i18ngrep.
>> >
>> > It's unclear what the '!' should negate in case of 'test_path_is_file
>> > ! file'.  What if 'file' does exists, but it's not a file but a
>> > directory, socket, fifo, or symlink?  'test ! -f file' returns success
>> > in these cases as well.
>> >
>> > OTOH, it's quite clear what the negation should mean in case of
>> > 'test_i18ngrep'.
>>
>> *Should* we make it better? Yeah sure, maybe. I'm just pointing out for
>> context to someone poking at this for the first time that now we
>> sometimes do "! foo <arg>" v.s. "foo <arg>" as "foo_is <arg>" and
>> "foo_not <arg>" and other times "foo [!] <arg>".
>>
>> So yeah, maybe we should improve things to disambiguate the cases you
>> mentioned, but right now e.g. "test_path_exists" and
>> "test_path_is_missing" are just "test -e" and "! test -e", respectively.
>
> I'm not sure why 'test_path_exists' exists, as I don't readily see a
> valid reason why a test should only be interested in whether the path
> exists, but but not whether it's a file or a directory.

In the general case the same reason we use "test -e". While the test would pass in all sorts of unexpected cases, those probably aren't plausible and we're just e.g. checking "did the thing create a file it'll create in XYZ mode?"....

> Luckily it
> haven't caught on, and it's only used twice in the whole test suite.

Well, we have some >100 "test -e" though ... :)

Show 23 quoted lines

> However, as shown above, the existend of 'test_path_is_missing' is
> very much justified.
>
>> The same caveats you've mentioned also apply to "test_i18ngrep" b.t.w.,
>> there we squash the 3x standard exit codes of grep[1] into a boolean,
>> and thus e.g. ignore the difference between <file> not matching an
>> <file> being a directory or whatever.
>
> I'm not sure I understand, 'test_i18ngrep' handles the missing file or
> not a file cases reasonably well:
>
>   expecting success:
>           test_i18ngrep ! foo nonexistent-file
>
>   error: bug in the test script: test_i18ngrep requires a file to read as the last parameter
>
> or
>
>   expecting success:
>           mkdir dir &&
>           test_i18ngrep ! foo dir
>
>   error: bug in the test script: test_i18ngrep requires a file to read as the last parameter

Yeah you're right, I didn't read it carefully enough and it does handle *that* particular case, but e.g. a grep of "file" where we can't read it due to a permission error is the same as "didn't contain the string".

I'm not saying that's a bad thing, but the opposite. We assume a certain base level of sanity, e.g. we do "test_must_fail <cmd>" only for git, but "! <cmd>" for everything else, even though e.g. the system "grep" may be segfaulting.

>>
>> 1. http://pubs.opengroup.org/onlinepubs/009695399/utilities/grep.html