From: Junio C Hamano <gitster@pobox.com>
Date: Fri, 17 May 2024 17:55:33 GMT
Subject: Re: Three t4150 tests does not work as expected
Message-ID: <xmqqcypkiaqy.fsf@gitster.g>
In-Reply-To: <ZkeJx97qdLoaC2H-@telcontar>

Marcel Telka <marcel@telka.sk> writes:

> All of them does something like:
>
> 	git show HEAD --format="%B" >actual
> 	grep -f actual expected

Wow, these commands look nonsensical in as many counts as there are
lines in there X-< [*].

It seems that they came from two patches, namely

 7c096b8d (am: support --empty=<option> to handle empty patches, 2021-12-09)
 9e7e41bf (am: support --allow-empty to record specific empty patches, 2021-12-09)

Most likely the author of these lines were confused and meant to use
test_cmp (which by the way should always compare actual against
expect, i.e. "test_cmp expect actual", because when the actual
output differs, that is the order of the diff we want to see in the
output to diagnose the breakage).  I especially suspect that the
author did not mean "read patterns from this file", but was confused
to think that 'f' implies "match literally" somehow (the correct
spelling of that fixed-strings option is -F but still it does the
wrong thing reading from the file---it does not even ensure that
all patterns listed in the file matches).

I am not saying (as I did not check) that doing just "grep -f A B"
-> "test_cmp B A" woudl make things work.  I do not think I trust
that these tests added by the same patch have a sane expectations
X-<.  The outcome these tests try to verify may also need to be
corrected.


[Footnote] 

 * To start with, any dashed option like --format=%B should come
   before a revision argument HEAD.  There are many other
   antipatterns in this test script, which should be cleaned up.