On Wed, Mar 10 2021, Junio C Hamano wrote:
> Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes:
>
>> Remove the test_external function(s) in favor of running the Perl
>> tests with a test_expect_success.
>> ...
>> My motivation for this is to eliminate a special case where things
>> that aren't test-lib.sh are going to produce TAP, for reasons that'll
>> be clear in subsequent commits.
>
> Puzzled.
>
>> .../netrc/t-git-credential-netrc.sh | 7 +-
>> t/README | 26 ------
>> t/t0202-gettext-perl.sh | 10 +--
>> t/t9700-perl-git.sh | 10 +--
>> t/test-lib-functions.sh | 89 +------------------
>> t/test-lib.sh | 42 ++++-----
>> 6 files changed, 28 insertions(+), 156 deletions(-)
>
> Reducing the number of lines is always good, but is this essentially
> losing what the commit that added test_external wanted to add?
Yes, I don't think streaming TAP from external sources is a problem
worth solving in the first place.
>> diff --git a/contrib/credential/netrc/t-git-credential-netrc.sh b/contrib/credential/netrc/t-git-credential-netrc.sh
>> index 07227d02287..28118a9e194 100755
>> --- a/contrib/credential/netrc/t-git-credential-netrc.sh
>> +++ b/contrib/credential/netrc/t-git-credential-netrc.sh
>> @@ -20,13 +20,10 @@
>> 'set up test repository' \
>> 'git config --add gpg.program test.git-config-gpg'
>>
>> - # The external test will outputs its own plan
>> - test_external_has_tap=1
>> -
>> export PERL5LIB="$GITPERLLIB"
>> - test_external \
>> - 'git-credential-netrc' \
>> + test_expect_success 'git-credential-netrc' '
>> perl "$GIT_BUILD_DIR"/contrib/credential/netrc/test.pl
>> + '
>>
>> test_done
>> )
>
> This is valid because we expect nobody runs this under tap?
No, because at this point we suppress the --verbose output under
HARNESS_ACTIVE (but see 6/7 later), and under non-verbose we now emit:
./t9700-perl-git.sh
ok 1 - set up test repository
ok 2 - use t9700/test.pl to test Git.pm
# passed all 2 test(s)
1..2
So that the stdout from t9700/test.pl doesn't screw up the test output
anymore.
>> diff --git a/t/t0202-gettext-perl.sh b/t/t0202-gettext-perl.sh
>> index a29d166e007..06a93b36790 100755
>> --- a/t/t0202-gettext-perl.sh
>> +++ b/t/t0202-gettext-perl.sh
>> @@ -17,11 +17,9 @@ perl -MTest::More -e 0 2>/dev/null || {
>> test_done
>> }
>>
>> -# The external test will outputs its own plan
>> -test_external_has_tap=1
>> -
>> -test_external_without_stderr \
>> - 'Perl Git::I18N API' \
>> - perl "$TEST_DIRECTORY"/t0202/test.pl
>> +test_expect_success 'run t0202/test.pl to test Git::I18N.pm' '
>> + perl "$TEST_DIRECTORY"/t0202/test.pl 2>stderr &&
>> + test_must_be_empty stderr
>> +'
>
> So t0202/test.pl would still produce output that would confuse
> whoever is reading our output as TAP, and it is OK? If the
> redirection discards its standard output to /dev/null [*], I would
> sort-of understand how this may work (we would have let the perl
> script to emit 13 "ok" or "not ok" to our output, but now we discard
> that and write just one our own "ok" or "not ok", depending on what
> comes out to the standard error stream (e.g. "# Looks like you
> failed...").
>
> But that is not what is going on. We'll let these 13 "ok" or "not ok"
> come out from the perl script and then add another on our own.
Yes, but just like we don't have stdout from any other program appear
directly in the output of ./tXXXX*.sh the TAP doesn't get to our stderr
either, i.e. our whole juggling around with FDs 1-4.
>> diff --git a/t/t9700-perl-git.sh b/t/t9700-perl-git.sh
>> index 102c133112c..574c57b17f1 100755
>> --- a/t/t9700-perl-git.sh
>> +++ b/t/t9700-perl-git.sh
>> @@ -50,11 +50,9 @@ test_expect_success \
> > git config --add test.pathmulti bar
>> '
>>
>> -# The external test will outputs its own plan
>> -test_external_has_tap=1
>> -
>> -test_external_without_stderr \
>> - 'Perl API' \
>> - perl "$TEST_DIRECTORY"/t9700/test.pl
>> +test_expect_success 'use t9700/test.pl to test Git.pm' '
>> + perl "$TEST_DIRECTORY"/t9700/test.pl 2>stderr &&
>> + test_must_be_empty stderr
>> +'
>
> Ditto. It seems that we are still letting the script, i.e. one of
> the "things that aren't test-lib.sh" to produce TAP anyway.
>
>
> [Footnote]
>
> * If we are truly somehow discarding these output that would be TAP
> (13 tests in 0202 and uncounted in 9700) from being shown (by e.g.
> redirecting output to /dev/null), it would be a regression for those
> who debug breakage found by these tests. They used to be told which
> one failed and how but now they don't. So I do not think that is a
> useful way to go, either.
You still get the full TAP output under --verbose for debugging, we just
don't consume it as TAP anymore by "prove" etc, i.e. once we get to 6/7:
$ GIT_TEST_TEE_STARTED=true HARNESS_ACTIVE=t ./t9700-perl-git.sh --verbose --tee
[...]
ok 1 - set up test repository
## expecting success of 9700.2 'use t9700/test.pl to test Git.pm':
perl "$TEST_DIRECTORY"/t9700/test.pl 2>stderr &&
test_must_be_empty stderr
\ok 2 - use Git;
\ok 3 - open repository
\ok 4 - config scalar: string
\ok 5 - config array: string
\ok 6 - config scalar: nonexistent
\ok 7 - config array: nonexistent
[...]
\ok 44 - cat_blob(outside): size
\ok 45 - unquote unquoted path
\ok 46 - unquote simple quoted path
\ok 47 - unquote escape sequences
\1..47
ok 2 - use t9700/test.pl to test Git.pm
# passed all 2 test(s)
1..2
Notice how it's escaped now.
So yes, we don't get the benefit of having e.g. prove/CI say that our
6th test failed anymore in this case.
I think that's a very small price to pay for the benefit of not having
to monkeypatch Perl's TAP emitting in this case to do the back & forth
munging to safe the TAP output I'm doing in 6/7, and only need to do in
one place because I'd gotten rid of this special-case earlier.
