Hi Eric,

On Wed, 19 Jun 2019, Johannes Schindelin wrote:

On Wed, 19 Jun 2019, Eric Sunshine wrote:

quoted

On Wed, Jun 19, 2019 at 2:19 AM Johannes Sixt [off-list ref] wrote:

quoted

Am 19.06.19 um 01:12 schrieb Eric Sunshine:

quoted

On Tue, Jun 18, 2019 at 8:24 AM Johannes Schindelin via GitGitGadget
[off-list ref] wrote:

quoted

        printf ".git\nfatal: Needed a single revision\n" >expect &&
-       test_cmp expect output.txt
+       sort <output.txt >output.sorted &&
+       test_cmp expect output.sorted

It was quite surprising to see this sorting only 'output' but not
'expect'. I see now that 'output' is already "sorted" (in that sense),
but it feels fragile. More robust would be to sort 'expect' as well:

    printf ".git\nfatal: Needed a single revision\n" | sort >expect &&

Following Dscho's recent objection elsewhere that tests tend to check
for much more than regressions, wouldn't it be logical to write these as

        grep -F .git" output.txt &&
        test_i18n_grep "Needed a single rev" output.txt

without an 'expect' file at all?

I considered suggesting that, as well, as being more obvious and less
fragile (with the exception that "Needed a single rev" isn't currently
localizable in builtin/rev-parse.c, so plain 'grep' instead of
'test_i18n_grep').

Interesting side note: I just realized that t6050-replace.sh does indeed
contain

	test_i18ngrep "Needed a single revision" err

so I wonder why that works.

If anybody has an answer, I'd be curious, but for now I want to focus on
this here patch series instead.

Ciao,
Dscho

Valid points all around, thank you so much!

The next iteration will have the two `grep`s instead,
Dscho