[SOLVED] bug while building glibc-2.27 with new gcc-8.1.1 and perl-5.28.0

If you try to rebuild glibc-2.27 as for SFS or in Slackware-current you get that message, not a good one:
--------------------------------
touch /tmp/glibc-2.27/build-glibc-2.27/manual/stamp-version
LANGUAGE=C LC_ALL=C makeinfo -P /tmp/glibc-2.27/build-glibc-2.27/manual/ --output=/tmp/glibc-2.27/build-glibc-2.27/manual/libc.info libc.texinfo
Unescaped left brace in regex is deprecated here (and will be fatal in Perl 5.32), passed through in regex; marked by <-- HERE in m/^\s+@([[:alnum:]][[:alnum:]\-]*)({ <-- HERE })?\s*/ at /usr/share/texinfo/Texinfo/Parser.pm line 5481.
Unescaped left brace in regex is deprecated here (and will be fatal in Perl 5.32), passed through in regex; marked by <-- HERE in m/^\s+@([[:alnum:]][[:alnum:]\-]*)({ <-- HERE })?\s*(\@(c|comment)((\@|\s+).*)?)?/ at /usr/share/texinfo/Texinfo/Parser.pm line 5485.
----------------------------------
It can't get to the end.
The problem is in texinfo, see the message: https://bugs.debian.org/cgi-bin/bugr...cgi?bug=898994

I've not tried to rebuild glibc, but I think those are just warnings: when I tried to build qemu I met them too but rebuilding texinfo with this patch made them go away.
that unfortunately hadn't solved the issue (makeinfo gets stuck) that I think is related to makeinfo/tex2any (a perl script) when used with perl 5.28.x to process specific files.

I tried the same patch, but as you said, it gets stuck at the same level:
-----------------------------
/bin/sh ../scripts/move-if-change /tmp/glibc-2.27/build-glibc-2.27/manual/version-tmp /tmp/glibc-2.27/build-glibc-2.27/manual/version.texi
touch /tmp/glibc-2.27/build-glibc-2.27/manual/stamp-version
LANGUAGE=C LC_ALL=C makeinfo -P /tmp/glibc-2.27/build-glibc-2.27/manual/ --output=/tmp/glibc-2.27/build-glibc-2.27/manual/libc.info libc.texinfo
----------------------------
it doesn't go the end, the last line stays for ever, for ever, for ever....

I had the same issue with building Qemu, it hangs on the document generating. I have bug report open with texinfo and already talking with a developer...

It suspect it to be a problem with texinfo in combination with the new Perl.

Kr
Mel

It appears that makeinfo hangs if the document includes @documentencoding UTF-8 but the calling locale is not UTF-8 (such as when the glibc build calls it with LANGUAGE=C LC_ALL=C). Removing the line "@documentencoding UTF-8" from libc.texinfo works around the problem, and I'll be uploading a FTBFS fix with this once I've verified that it works here with a full glibc build. Another workaround would be to have the build call makeinfo with LANGUAGE=C.UTF-8 LC_ALL=C.UTF-8.

I'm not exactly sure where the blame lies with this one. A genuine fix would be welcome, but this will do for now.

Originally Posted by volkerdi It appears that makeinfo hangs if the document includes @documentencoding UTF-8 but the calling locale is not UTF-8 (such as when the glibc build calls it with LANGUAGE=C LC_ALL=C). Removing the line "@documentencoding UTF-8" from libc.texinfo works around the problem

I can confirm that this workaround is effective also for qemu-doc.texi, thanks Pat!

I have passed the information on to the texinfo developer assuming it is a texinfo issue...

for the sake of completness, during the qemu build, just after having processed qemu-doc.texi, this happens (still after upgrading to the newer perl)

perl -Ww -- /tmp/SBo/qemu-2.12.0/scripts/texi2pod.pl -I docs -I . -I . "-DVERSION=2.12.0" qemu-doc.texi qemu.1.pod && pod2man --utf8 --section=1 --center=" " --release=" " qemu.1.pod > qemu.1                                                              
qemu.1.pod around line 95: Non-ASCII character seen before =encoding in 'Schtz.'. Assuming UTF-8
POD document had syntax errors at /usr/bin/pod2man line 71.
make: *** [/tmp/SBo/qemu-2.12.0/rules.mak:385: qemu.1] Error 255
make: *** Deleting file 'qemu.1'

Originally Posted by ponce for the sake of completness, during the qemu build, just after having processed qemu-doc.texi, this happens (still after upgrading to the newer perl) perl -Ww -- /tmp/SBo/qemu-2.12.0/scripts/texi2pod.pl -I docs -I . -I . "-DVERSION=2.12.0" qemu-doc.texi qemu.1.pod && pod2man --utf8 --section=1 --center=" " --release=" " qemu.1.pod > qemu.1 qemu.1.pod around line 95: Non-ASCII character seen before =encoding in 'Schtz.'. Assuming UTF-8 POD document had syntax errors at /usr/bin/pod2man line 71. make: *** [/tmp/SBo/qemu-2.12.0/rules.mak:385: qemu.1] Error 255 make: *** Deleting file 'qemu.1'

Yes, I noticed this too, but when I changed the make file and replaced LC_ALL=C with "LANGUAGE=C.UTF-8 LC_ALL=C.UTF-8", it worked perfectly.

Kr
Mel

yes, but if for qemu this

sed -i "s|LC_ALL=C|LC_ALL=C.UTF-8|" Makefile

solves both the issues, with makeinfo (shipped with texinfo) and with pod2man (provided by perl itself), then my bet is for perl 5.28.0 to blame.

double post

Hi,

after talking with one of the developers we tried to escape the second bracket too... this seems to solve the issue, it did for me:

--- /usr/share/texinfo/Texinfo/Parser.pm.bak	2018-06-28 07:05:26.046994398 +0200
+++ /usr/share/texinfo/Texinfo/Parser.pm	2018-06-28 07:06:45.512993300 +0200
@@ -5478,11 +5478,11 @@
     }
   } elsif ($command eq 'clickstyle') {
     # REMACRO
-    if ($line =~ /^\s+@([[:alnum:]][[:alnum:]\-]*)({})?\s*/) {
+    if ($line =~ /^\s+@([[:alnum:]][[:alnum:]\-]*)(\{\})?\s*/) {
       $args = ['@'.$1];
       $self->{'clickstyle'} = $1;
       $remaining = $line;
-      $remaining =~ s/^\s+@([[:alnum:]][[:alnum:]\-]*)({})?\s*(\@(c|comment)((\@|\s+).*)?)?//;
+      $remaining =~ s/^\s+@([[:alnum:]][[:alnum:]\-]*)(\{\})?\s*(\@(c|comment)((\@|\s+).*)?)?//;
       $has_comment = 1 if (defined($4));
     } else {
       $self->line_error (sprintf($self->__(

I hope I created the patch correctly..

EDIT: it worked once....

more info on the matter: Gavin, on the bug report, suggested to run makeinfo with TEXINFO_XS=omit and when using that (disabling texinfo perl extensions) build goes smooth.

Yeah I noticed the same. Responded to his email too...

Lets see where this go, but I agree that Perl is probably the culprit.

Kr
Mel

Answer from the dev

When one installs a new version of Perl, isn't one supposed to build the XS extensions anew, against the newly installed Perl library? Was that done in this case?

Can anyone comment on this? I would assume so, but I am unsure...