libgcc @7.2.0: fatal error: bits/regex_scanner.tcc: No such file or directory

[cjones051073 (Chris Jones)]

Running newly release macOS 10.13. Xcode and CLI tools fully updated.

building libgcc gives

:info:build                  from /opt/local/var/macports/build/_Users_chris_Projects_MacPorts_ports_lang_gcc7/libgcc/work/gcc-7.2.0/libstdc++-v3/include/precompiled/stdc++.h:110:
:info:build /opt/local/var/macports/build/_Users_chris_Projects_MacPorts_ports_lang_gcc7/libgcc/work/build/x86_64-apple-darwin17/libstdc++-v3/include/bits/regex_scanner.h:272:10: fatal error: bits/regex_scanner.tcc: No such file or directory
:info:build  #include <bits/regex_scanner.tcc>

Full log attached.

Port has no maintainer, so not sure who to cc. Looks similar though to ticket:54829

[ryandesign (Ryan Carsten Schmidt)]

Supersedes #54400.

[cjones051073 (Chris Jones)]

There appears to be a non-reproducible nature to the failure. I have run the build a few times now and each time the exact error is slightly different. Its in roughly the same place, and relates to an "bits/XYZ" include, but not exactly the same include each time. For instance

:info:build /opt/local/var/macports/build/_Users_chris_Projects_MacPorts_ports_lang_gcc7/libgcc/work/build/x86_64-apple-darwin17/libstdc++-v3/include/string:41:10: fatal error: bits/allocator.h: No such file or directory
:info:build  #include <bits/allocator.h>
:info:build           ^~~~~~~~~~~~~~~~~~

Could this be an APFS issue (the machine I am using was updated to the new filesystem). I ran checks with the betas version prior to the release, in a VM, and the only difference I can think of is that VM was not using APFS but the old file system...

[ryandesign (Ryan Carsten Schmidt)]

The High Sierra buildbot worker will be set up with HFS+ so we can see if it succeeds there.

[cjones051073 (Chris Jones)]

I have a number of my own VMs, from 10.6 onwards that I use for testing. I am in the process of cloning the macOS 10.12 one to a new 10.13 instance, so I will also be able to test this...

[jwhowarth]

I believe you are running into ​https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81797 which occurs on APFS volumes. The issue appears be the fine granularity of the new filesystem breaking the time stamping scheme used in the install-headers section of libstdc++-v3/include/Makefile.in (which is based on the existence of directories and not their population state). The workaround is to use the change...

diff -uNr gcc-7.2.0.orig/libstdc++-v3/include/Makefile.in gcc-7.2.0/libstdc++-v3/include/Makefile.in
--- gcc-7.2.0.orig/libstdc++-v3/include/Makefile.in     2017-07-25 14:05:07.000000000 -0400
+++ gcc-7.2.0/libstdc++-v3/include/Makefile.in  2017-09-02 12:22:08.000000000 -0400
@@ -1764,6 +1764,8 @@
 @GLIBCXX_HOSTED_TRUE@install-data-local: install-headers
 @GLIBCXX_HOSTED_FALSE@install-data-local: install-freestanding-headers
 
+.NOTPARALLEL: install-headers
+
 # This is a subset of the full install-headers rule.  We only need <ciso646>,
 # <cstddef>, <cfloat>, <limits>, <climits>, <cstdint>, <cstdlib>, <new>,
 # <typeinfo>, <exception>, <initializer_list>, <cstdalign>, <cstdarg>,

[cjones051073 (Chris Jones)]

Thanks for that bug report. Looks like exactly what I am seeing.

I'll see if I can roll an update to add the workaround you mention above.

[kencu (Ken)]

note that this might start to show up on other gcc* builds with APFS file systems as well. disabling parallel builds would be the easy but tedious solution.

[cjones051073 (Chris Jones)]

Indeed. The gcc bug tracker above has a report that gcc6 at least was affected.

Assuming my ongoing libgcc (gcc7) build works I was going to look to making a similar patch for gcc6 and gcc5 and then submit a pull request for these. I am not sure if anything older is needed (or even builds) on macOS 10.13...

[kencu (Ken)]

there are some other gcc-like ports as well in MacPorts that will likely be affected too. upc for one...

[cjones051073 (Chris Jones)]

Lets start with the gcc ports... These are dependencies for a number of others so first fix these and then move on...

[cjones051073 (Chris Jones)]

OK, libgcc and gcc7 built fine with the above patch. I'll submit an update for this shortly.

Unfortunately a similar fix for gcc6 does not seem to work. See attached build log. If anyone has any ideas on this I would like to hear ?

Chris

[cjones051073 (Chris Jones)]

see ​https://github.com/macports/macports-ports/pull/834

[jeremyhu (Jeremy Huddleston Sequoia)]

#54829