Re: [PATCH v2 19/27] gitattributes doc: document multi-line userdiff patterns

On Mon, Feb 15 2021, Johannes Sixt wrote:

Am 15.02.21 um 16:44 schrieb Ævar Arnfjörð Bjarmason:

quoted

diff --git a/t/t4018/custom.sh b/t/t4018/custom.sh
index 72d38dad68..30df13d8b2 100755
--- a/t/t4018/custom.sh
+++ b/t/t4018/custom.sh

@@ -111,3 +111,53 @@ ChangeMe
    baz
  EOF_TEST
+
+test_expect_success 'custom: setup negation syntax, ! is magic' '
+	git config diff.custom.xfuncname "!negation
+line"
+'
+
+test_diff_funcname 'custom: negation syntax, ! is magic' \
+	8<<\EOF_HUNK 9<<\EOF_TEST
+line
+EOF_HUNK
+line
+!negation
+
+ChangeMe
+
+baz
+EOF_TEST
+
+test_expect_success 'custom: setup negation syntax, use [!] to override ! magic' '
+	git config diff.custom.xfuncname "[!]negation
+line"
+'
+
+test_diff_funcname 'custom: negation syntax, use [!] to override ! magic' \
+	8<<\EOF_HUNK 9<<\EOF_TEST
+!negation
+EOF_HUNK
+line
+!negation
+
+ChangeMe
+
+baz
+EOF_TEST
+
+test_expect_success 'custom: setup captures in multiple patterns' '
+	git config diff.custom.xfuncname "!^=head
+^format ([^ ]+)
+^sub ([^;]+)"
+'
+
+test_diff_funcname 'custom: captures in multiple patterns' \
+	8<<\EOF_HUNK 9<<\EOF_TEST
+foo
+EOF_HUNK
+sub foo;
+=head1
+ChangeMe
+
+EOF_TEST

This test would not catch a regression. You must leave a line between
the candidate-that-must-not-be, =head1, and ChangeMe, otherwise,
=head1 is never tested against the negation pattern.

Or did you change the diff invocation in an earlier patch such that it
does not emit context lines?

quoted

diff --git a/t/t4018/perl.sh b/t/t4018/perl.sh
index ac8fff7417..2952483a2c 100755
--- a/t/t4018/perl.sh
+++ b/t/t4018/perl.sh

@@ -76,3 +76,19 @@ sub RIGHT
  	print "ChangeMe\n";
  }
  EOF_TEST
+
+
+test_expect_success 'custom: setup config overrides built-in patterns' '
+	git config diff.perl.xfuncname "!^=head
+^[^ ]+.*"
+'
+
+test_diff_funcname 'custom: config overrides built-in patterns' \
+	8<<\EOF_HUNK 9<<\EOF_TEST
+sub foo;
+EOF_HUNK
+sub foo;
+=head1
+ChangeMe
+
+EOF_TEST

Same here.

Well spotted. Both of these test work for their advertised
purpose. I.e. "we can override the perl built-in config" and "capture
count gets reset", but you're right that the =head1 negation rule is
redundant at this point in the series.

In 23/27 we start testing with -U0, at which point it starts
working. But of course a better fix is to just add an empty line here...