I've been looking at this a little more.

This is what happens when I build in a Debian VM:
The first time that autogen.sh is run, autopoint copies everything into the= intl/ directory.
The second time that autogen.sh is run, autopoint only copies intl/plural.c= and po/Makefile.in.in.
So, since plural.c is newer than plural.y, make wont regenerate plural.c from plural.y and the new plural.c is used. The two plural.c files=20 aren't the same, they're generated from two different version of Bi= son.=20 From the diff:

-/* A Bison parser, made by GNU Bison 2.7. */ (build by make from plura= l.y)
+/* A Bison parser, made from plural.y
+ by GNU bison 1.35. */ (2nd invocation overwrites built version with th= is, this version works)

I'm not sure where the second plural.c file is coming from (or the f= irst for that matter), I can't find it anywhere on my system. Is it dyn= amically generated? Bison on the Debian VM is version 3.0.2.

The= version of gettext that gets included in the distribution is determined by= the AM_GNU_GETTEXT_VERSION macro in config= ure.ac. This is read by autopoint when autogen.sh is executed, and the = files for whatever the requested version of gettext is are copied into intl= /.

Presently, we're calling out version 0.14 in configure.ac. I tried bumping gettext to version 0.= 19.3 (what's installed on the VM) and make distcheck succeeds the first= time. This version was released in 2014-10, so it's a little over two = years old. I tried with version 0.18, and that also fails the same way as w= ith version 0.14.

So, another solution is to bump the version of gett= ext.

Any thoughts on this? I know we shouldn't arbitrarily bu= mp external version requirements, and I haven't heard any bug reports o= n the internationalization front, so, we would only be doing this so that m= ake distcheck builds cleanly, so maybe we should just not worry about it. <= br>

The --with-included-gettext option is used to configure in the ma= ke distcheck target. So, I imagine that most people are not specifying this= option and using their own external version when they build, which is prob= ably newer anyway (but I could be wrong). Does anyone know why automake inc= ludes this option in the distcheck target?

--Chad

=20 =20 =20

On = Mon, Jan 2, 2017 at 11:16 PM, Dan McMahill (dan AT mcmahill DOT net) [via ge= da-user AT delorie DOT com] <geda-user AT delorie DOT com> wrote:<= br>

On 1/1/2017 5:26 PM, Cha= d Parker (par= ker DOT charles AT gmail DOT com) [via geda-user AT delorie DOT com] wrote:

All-

While working on preparing a release, I've run into a bug that I could<= br> use some advice on.

Make distcheck fails in the intl/ directory while trying to build
plural.o (based on plural.c from plural.y). I don't quite understand what's behind the error, but apparently its a bug related to Bison 3. I've posted a few references at the end of this bug report that
references the errors:
https://bugs.launchpad.net/pcb/+bug/1496105
although, the solution of running the sequence twice doesn't appear to<= br> work for me.

When executing the make distcheck, configure is executed with the
--with-included-gettext option. Removing this option allows the
distcheck to succeed. Does anyone know if there's a particular reason for this? Removing this option is the easiest solution I think, if it
doesn't cause some other problems. There does appear to be an alternate=
solution, but it is more complicated.

That is one of the details of gettext that I don't clearly understand.<= br>
When running the 'make distcheck' twice and getting it to work, it = would be interesting to see the difference between the source tree before a= nd after the first one.

If we have some dependencies on particular bison behaviors or versions, or = also requiring flex instead of lex, we should capture that in the documenta= tion and some configure checks.

I'm curious if you had any other issues.=C2=A0 I'm failing 'mak= e distcheck' because of some broken stuff in src/Makefile.am.=C2=A0 I&#= 39;ll start a different thread for that.

-Dan