netpbm: parallel build fails

[ryandesign (Ryan Carsten Schmidt)]

With netpbm 10.26.58 and MacPorts trunk (@42849) on Mac OS X 10.4.11 Intel, the parallel build now fails:

--->  Fetching netpbm
--->  Verifying checksum(s) for netpbm
--->  Extracting netpbm
--->  Applying patches to netpbm
--->  Configuring netpbm
--->  Building netpbm
Error: Target org.macports.build returned: shell command " cd "/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58" && nice -n 10 gnumake -j2  messages=yes " returned error 2
Command output: rm -f nstring.h
rm -f mallocvar.h
ln -s /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib/util/nstring.h nstring.h
ln -s /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib/util/mallocvar.h mallocvar.h
gnumake -C /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib/ -f /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib/Makefile \
    SRCDIR=/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58 BUILDDIR=/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58 libnetpbm.dylib 
gnumake -C /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/buildtools/ -f /mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/buildtools/Makefile \
    SRCDIR=/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58 BUILDDIR=/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58 libopt 
/usr/bin/gcc-4.0 -c -I/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib -I. -DNDEBUG -I/mp/include -O3 -fno-common \
    -o libpm.o libpm.c
/usr/bin/gcc-4.0 -c -I/mp/include -O3 -o libopt.o \
-DSHLIBPREFIXLIST="\"lib\"" \
            \
        libopt.c
In file included from pm.h:16,
                 from libpm.c:42:
pm_config.h:487: error: redefinition of typedef 'uint32n'
pm_config.h:178: error: previous declaration of 'uint32n' was here
pm_config.h:488: error: redefinition of typedef 'int32n'
pm_config.h:179: error: previous declaration of 'int32n' was here
In file included from pm.h:16,
                 from libpm.c:42:
pm_config.h:623: error: redefinition of typedef 'pm_filepos'
pm_config.h:314: error: previous declaration of 'pm_filepos' was here
gnumake[3]: *** [libpm.o] Error 1
gnumake[2]: *** [/mp/var/macports/build/_Users_rschmidt_macports_dports_graphics_netpbm/work/netpbm-10.26.58/lib/libnetpbm.dylib] Error 2
gnumake[2]: *** Waiting for unfinished jobs....
/usr/bin/gcc-4.0 -o libopt -L/mp/lib libopt.o
gnumake[1]: *** [pgm/all] Error 2
gnumake: *** [converter/all] Error 2

Error: Unable to upgrade port: 1

It succeeds if I turn off the parallel build.

[mas@…]

Did we not see this problem a few months ago? It seems mcalhoun enabled parallel build again yesterday.

For the record, it works on my 10.5.5 machine.

I'm disabling parallel build for the time being (r43078) and adding a note about the problem to the Portfile.

Both of you, feedback on how to handle the problem would be appreciated. Personally, I don't think MacPorts is all too resource observant, so we could probably live with parallel build turned off. It's better to have a working Portfile than to have a fast one that only works for some people.

[blb@…]

This brings up the question, how do people try to verify parallel is valid? The few times I've done so I usually try building the port several times, since race conditions are a bitch. Even then, you may not catch the issue, or it could be platform or OS version dependent...

[MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)]

While searching through the history of netpbm to see why parallel build was turned off in the first place,
I was unable to reproduce the error in #16560.
I assumed it had been fixed in one of the subsequent version updates.
It now seems to be Tiger specific.

Testing is a problem.
I would venture to guess that most of use only have one OS version to test on.
For netpbm, at least, we have documentation that parallel_build has historically been a problem on Tiger.

Unfortunately, I have not useful suggestions on how this situation could be improved.

[MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)]

Since netpbm no longer tries to parallel build, can this ticket be closed?

[ryandesign (Ryan Carsten Schmidt)]

I informed netpbm's author, Bryan Henderson, of the problem again.

[ryandesign (Ryan Carsten Schmidt)]

Bryan explained that he had fixed the problem in his "Stable", "Advanced" and "Development" branches, but not in the "Super Stable" branch, which is the one the MacPorts netpbm port seems to be following, perhaps because "Super Stable" is the only branch for which netpbm releases source tarballs; the other branches are available only from their Subversion repository.

Sorry, I must not have realized you were using the Super Stable (10.26)
series.  I fixed the parallel make only in Stable, Advanced, and Development.
I don't think this change is a good idea in Super Stable because a) the make
files are rather different there; b) there probably is other work that would
have to be done to make parallel make truly work; and c) parallel make adds
little value.

But if you want to stay with 10.26 and have parallel make, you could try
applying the same change locally.  It's Subversion Revision 733 of 'trunk',
which is this:

He then provided a diff of his revision 733, which I've attached here. We can either try to apply that and see if it works (he thinks it might not), or we can just leave parallel building off until the Stable branch becomes the Super Stable branch. Or we can try to switch the port to use the Stable branch now.

[ryandesign (Ryan Carsten Schmidt)]

r45211: added note to portfile about the situation

[(none)]

Milestone Port Bugs deleted

[ryandesign (Ryan Carsten Schmidt)]

netpbm is now on the "advanced" branch, on which parallel building should be fine. So I reenabled it in r105291.

[cooljeanius (Eric Gallager)]

Cc Me!

[blair (Blair Zajac)]

In r112198 I updated netpbm to 10.64.01. However, testing it on a dual-core system, I couldn't get the package to compile after three port -v clean netpbm; port -v upgrade --enforce-variants outdated. The first attempt after removing parallel build option it succeed. So in r112199 I disabled the parallel builds.

[kencu (Ken)]

I had this issue again yesterday, upgrading netpbm to @10.77.03_2 on an older Dual G5 running Tiger.

The build appears to hang in mid-build, with no error message.

Disabling parallel builds again allows it to continue to completion.

Just leaving this cookie here for people who may stumble across this issue even now.

[ryandesign (Ryan Carsten Schmidt)]

Ken, could you attach your main.log file from that hang?

[kencu (Ken)]

I killed that build after a period of time hung, btw - ergo the interrupts at the end. K.