Created attachment 184861 [details] Patch to delete include-fixed directory during post-stage gcc5 (r444860), gcc5-devel (r444563), gcc6 (r445313), and gcc6-devel (r445310) were recently patched to remove the headers created by GCC's fixincludes process by deleting the include-fixed directory during post-stage. gcc6-aux has not been patched yet (since it currently has no maintainer), so the attached patches do the same for gcc6-aux. If the BOOTSTRAP option is set, the headers are deleted only in the bootstrap tarball created by the Makefile. The attached patch deletes the headers before the tarball is created. For more information on the background of this issue, please see the following threads: https://lists.freebsd.org/pipermail/freebsd-stable/2017-April/087087.html https://lists.freebsd.org/pipermail/freebsd-stable/2017-June/087306.html
Created attachment 184862 [details] PORTREVISION bump
I had to move the RM line above the preceding command (creating TMPPLIST) for this to work (failed on packaging otherwise.)
Created attachment 184871 [details] Revised Makefile patch to delete include-fixed directory during post-stage Yeah, sorry about that. Uploaded the wrong patch for the Makefile. I've provided a corrected patch. This one definitely builds withouth plist problems.
Thank you David. Could you confirm this change (and thus the port) passes QA please (portlint & poudriere in particular). Also, if both patches are required (ie; one is not obsolete), combine them into a single diff. CC GCC maintainer (can think of noone more relevant) in case they can assist landing this given this port has no maintainer. Also, assuming the quarterly version is also affected, please confirm
(In reply to Kubilay Kocak from comment #4) > CC GCC maintainer (can think of noone more relevant) in case they can > assist landing this given this port has no maintainer. I'll help (as long as I won't be held accountable for this port going forward in turn ;-).
(In reply to Gerald Pfeifer from comment #5) I read that as, 'this port needs a maintainer' Thank you for the assist!
bug 221237 reports what appear to be very similar results/symptoms/errors but in gcc5. Please confirm and remove dependency (after creating a separate issue for it) if not. If so, please update this issue summary to reflect multiple gcc ports are affected
The description for bug #221237 mentions the problem is with 2017Q3. Gerald committed the fix for gcc5 on July 2, which might have been too late to be included in 2017Q3, and the fix was not MFH to 2017Q3 as of the time I post this. I'll set aside some time this weekend to do some portlint & poudriere testing -- I'm a bit of a newbie when it comes to patching Makefiles, and I've got some learning to do!
(In reply to David Z. from comment #8) Since that commit was made before this issue was created, its best for issue tracking purposes that we treat them separately, even though both bugs represent the same logical issue. Normally one would want to identify all ports affected by the reported issue, and have all commits/merges reference to that one issue (by including PR: references in the commit log message). Therefore, undo blocking bug 221237, leaving this issue scoped to gcc6-aux. It is however, quite possible other gcc ports are still affected either in head or quarterly (I haven't looked). If this is the case, a separate issue should be created (With this one in its "See Also:" field) listing all other affected gcc ports (excluding gcc5/quarterly and gcc6-aux/head&quarterly, covered by the existing two bugs) so that they can also be sorted.
(In reply to David Z. from comment #8) > The description for bug #221237 mentions the problem is with 2017Q3. Gerald > committed the fix for gcc5 on July 2, which might have been too late to be > included in 2017Q3, and the fix was not MFH to 2017Q3 as of the time I post this. You're right, this was not in the tree when 2017Q3 branched. I now backported the change a few hours ago.
A commit references this bug: Author: gerald Date: Sat Sep 23 18:25:42 UTC 2017 New revision: 450463 URL: https://svnweb.freebsd.org/changeset/ports/450463 Log: Always remove headers being created by GCC's fixincludes machinery from the installation / packaging, not just when bootstrapping GCC (with option BOOTSTRAP on). This avoids breakage when FreeBSD's headers are changing after the build. PR: 221111 Submitted by: David Z. <daz@hyperchronos.org> Changes: head/lang/gcc6-aux/Makefile head/lang/gcc6-aux/Makefile.version
Sorry David, somehow I failed to make the actual commit back then. Can you please confirm this is good? (Also, lang/gcc5-aux is now marked deprecated, but I'm thinking to apply the same change there.)