On 18.06.21 16:41, Jason Merrill wrote:
* Being able to specify the PR numbers on the command line in addition
(currently, they are only extracted from the testsuite patches)
This bit seems unnecessary to me, since we want the commit to include
tests that identify the PR.
I full hearty disagree!
Martin Sebor's patch to extract the PR number from the testcase
filename, as an alternative to a comment, should be enough.
Regarding the PR from the filename, are we only talking about
new files – or about modifications to old files?
For new files, together with -p it seems to be useful; for
old files, I think there will be more spurious/false PRs than
useful/right PRs.
Looking through the last few commits, in all of the following
cases the -b would be useful — and I do regard the commits
messages are sensible:
* cc9c94d43dcfa98436152af9c00f011e9dab25f6 PR libstdc++/100387
added testsuite/25_algorithms/minmax/constrained.cc (etc.)
but file neither has a pr... name nor contains a PR line
(BTW: The fix is not primarily for the PR but fixes it as,
a side effect; hence, not having a PR line makes sense)
* 870b674f72d4894b94efa61764fd87ecec29ffde
'ranger' - removes a datastructure (no testcase but also
does not make sense here)
* 17a4bee01c3b29c5ccdd39f34384521e5d44135b
Fixes an ICE for one testcase (but of course does not add
a new testcase or modify an existing one.)
* 76e990fd211cbb20bf74ce074eb8b2d7b096d3b7
fix a bootstrap issue (found with the GCC 11 backport).
Those are all commits between last Friday and today. Hence, I do
believe it makes sense to be able to specify the PR on the
command line — besides obtaining it from file names.
Tobias
-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstrasse 201, 80634 München
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Frank
Thürauf
