On Wed, Mar 27, 2019 at 03:10:42PM +0100, Ævar Arnfjörð Bjarmason wrote:

On Wed, Mar 27 2019, SZEDER Gábor wrote:

quoted

On Wed, Mar 27, 2019 at 12:21:55PM +0100, Ævar Arnfjörð Bjarmason wrote:

quoted

On Wed, Mar 27 2019, SZEDER Gábor wrote:

quoted

On Wed, Mar 27, 2019 at 11:09:18AM +0100, Ævar Arnfjörð Bjarmason wrote:

quoted

quoted

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?"....

When I look at a test using 'test -e' I can't readily tell what that
reason might have been, i.e. whether there really was a compelling
reason for using 'test -e' instead of 'test -(f|d|...), or the dev
writing the test was just "meh, we could let it slide".

quoted

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 ... :)

Hopefully they will be turned into 'test_path_is_file' or
'test_path_is_dir' instead of 'test_path_exists'...  eventually, maybe
a decade or two worth of GSOC microprojects later... :)

quoted

However, as shown above, the existend of 'test_path_is_missing' is
very much justified.

quoted

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.

That's kind of the point: this is not about the sanity of the system,
but rather about the sanity of the developers :)

'test_i18ngrep' does handle that particular case, because,
unfortunately, paths do get misspelled (or downright forgotten!) every
now and then, but 'test_expect_failure', '! cmd', a pipe downstream,
or some other circumstances can still hide those errors.  I fixed
several issues with bogus filenames in the past.

However, since we assume that the system we are testing on is sane,
there is indeed not much point in preparing for cases that could only
go wrong if the system was not sane, e.g. the lack of permissions to
read a just written file.  I don't recall that we had any such
permission-related issues in the past (but I didn't actually check).
Having said that, I agree that checking whether 'grep' exited with 2
would have been more robust.


And these are the cases where using test helper functions that know
about the negation like 'test_path_is_missing', 'test_must_be_empty',
or 'test_i18ngrep !' instead of '! test -(f|d|s|...)' etc. starts to
pay off, as they can protect us from our own mistakes.  Case in point:
the last remaining '! test -s' in our test suite is definitely wrong.

So I think if such a helper function is available, then we should use
it, and we might want to consider adding even more, e.g. we have a
couple of instances of '! test_cmp', '! has_cr'.  And these functions
don't have to be perfect and catch every case that could possibly go
wrong, they just need to do a reasonably good job at catching those
cases that would be our own fault if they were to go wrong.

quoted

quoted

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