Junio C Hamano <gitster@xxxxxxxxx> writes:
> Eric Sunshine <sunshine@xxxxxxxxxxxxxx> writes:
>
>> That's very reminiscent of [1]. Although, unlike [1], the output
>> presented to the user in this case is (I suppose) less likely to be
>> messed up; only the combined captured output is probably affected. So,
>> capturing stdout and stderr separately would indeed be a good idea.
>
> Hmph, something along this line?
>
> It loses to capture how the output should be intermixed, which is
> essential to validate what the end-user should see. As we can see
> in the attached patch, we cannot express that "Unknown ..." should
> come in between two "Stage addition?" questions, which is a downside.
>
> Between adding fflush() before err() writes, and updating err() to
> write to the standard output stream, I am in favor of the latter for
> its simplicity (of the mental model of the resulting code, not of
> the patch that is required to do so).
The latter, which I claimed to prefer, would look like this.
The idea is to perform the interactive session over the standard
output (and the standard input). For that, we teach err() to use
the standard output and have a few fprintf() to also call err().
A few tests expect certain messages to appear on the standard error
stream, which needed adjusting.
I know the previous one "fixes" the CI job at Windows, but I haven't
tried this yet.
diff --git c/add-patch.c w/add-patch.c
index 7be142d448..c28ad380ed 100644
--- c/add-patch.c
+++ w/add-patch.c
@@ -293,10 +293,10 @@ static void err(struct add_p_state *s, const char *fmt, ...)
va_list args;
va_start(args, fmt);
- fputs(s->s.error_color, stderr);
- vfprintf(stderr, fmt, args);
- fputs(s->s.reset_color, stderr);
- fputc('\n', stderr);
+ fputs(s->s.error_color, stdout);
+ vfprintf(stdout, fmt, args);
+ fputs(s->s.reset_color, stdout);
+ fputc('\n', stdout);
va_end(args);
}
@@ -1326,7 +1326,7 @@ static int apply_for_checkout(struct add_p_state *s, struct strbuf *diff,
err(s, _("Nothing was applied.\n"));
} else
/* As a last resort, show the diff to the user */
- fwrite(diff->buf, diff->len, 1, stderr);
+ fwrite(diff->buf, diff->len, 1, stdout);
return 0;
}
@@ -1780,9 +1780,9 @@ int run_add_p(struct repository *r, enum add_p_mode mode,
break;
if (s.file_diff_nr == 0)
- fprintf(stderr, _("No changes.\n"));
+ err(&s, _("No changes."));
else if (binary_count == s.file_diff_nr)
- fprintf(stderr, _("Only binary files changed.\n"));
+ err(&s, _("Only binary files changed."));
add_p_state_clear(&s);
return 0;
diff --git c/t/t3701-add-interactive.sh w/t/t3701-add-interactive.sh
index 482d5c117e..a315ec99a3 100755
--- c/t/t3701-add-interactive.sh
+++ w/t/t3701-add-interactive.sh
@@ -43,17 +43,17 @@ force_color () {
}
test_expect_success 'warn about add.interactive.useBuiltin' '
- cat >expect <<-\EOF &&
+ cat >expect.error <<-\EOF &&
warning: the add.interactive.useBuiltin setting has been removed!
See its entry in '\''git help config'\'' for details.
- No changes.
EOF
for v in = =true =false
do
- git -c "add.interactive.useBuiltin$v" add -p >out 2>actual &&
- test_must_be_empty out &&
- test_cmp expect actual || return 1
+ git -c "add.interactive.useBuiltin$v" add -p >actual 2>error &&
+ echo "No changes." >expect &&
+ test_cmp expect actual &&
+ test_cmp expect.error error || return 1
done
'
@@ -348,13 +348,13 @@ test_expect_success 'different prompts for mode change/deleted' '
test_expect_success 'correct message when there is nothing to do' '
git reset --hard &&
- git add -p 2>err &&
- test_grep "No changes" err &&
+ git add -p >out &&
+ test_grep "No changes" out &&
printf "\\0123" >binary &&
git add binary &&
printf "\\0abc" >binary &&
- git add -p 2>err &&
- test_grep "Only binary files changed" err
+ git add -p >out &&
+ test_grep "Only binary files changed" out
'
test_expect_success 'setup again' '
