Created attachment 376276 [details] emerge --info make[2]: Entering directory '/var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/libraries/gc' /usr/bin/install -c -d /var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/BUILD/tarfiles error: for the third-party library or program source "gc" the source code is not present in the file "/var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/BUILD/tarfiles/gc-7.2d.tar.gz" so either download a "fat" tar file of the Macaulay2 source code or rerun the Macaulay2 "configure" command with the added option "--enable-download" to enable automatic downloading of the source code over the internet ../Makefile.library:168: recipe for target '/var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/BUILD/tarfiles/gc-7.2d.tar.gz' failed make[2]: *** [/var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/BUILD/tarfiles/gc-7.2d.tar.gz] Error 1 make[2]: Leaving directory '/var/tmp/portage/sci-mathematics/Macaulay2-1.6/work/M2-release-1.6-stable-20130514/M2/libraries/gc' Makefile:7: recipe for target 'fetch-in-gc' failed make[1]: *** [fetch-in-gc] Error 2 Earlier, from configure: checking for library containing GC_free... -lgc checking gc/gc.h usability... yes checking gc/gc.h presence... yes checking for gc/gc.h... yes checking atomic_ops.h usability... yes checking atomic_ops.h presence... yes checking for atomic_ops.h... yes checking whether libgc is recent enough... no, version at least 7.2 is required (gc will be built from downloaded sources) I have dev-libs/boehm-gc-7.4.0 installed, so there is something wrong with the autodetection in configure. Looking at the config.log: configure:9685: checking whether libgc is recent enough configure:9706: x86_64-pc-linux-gnu-gcc -o conftest -march=amdfam10 -ggdb -pipe -O0 -I/usr/include/frobby -DNDEBUG -Wl,--as-needed conftest.c -lgc -lmpfr -lfrobby -lpari -lmpirxx -lmpir -lreadline -lncurses -lpthread -ldl -L/usr/lib64/blas/reference -llapack -lblas >&5 […]/temp/ccad0k5A.o: In function `main': […]/M2/conftest.c:105: undefined reference to `GC_version' […]/M2/conftest.c:106: undefined reference to `GC_version' […]/M2/conftest.c:107: undefined reference to `GC_version' collect2: error: ld returned 1 exit status configure:9706: $? = 1 configure: program exited with status 1 configure: failed program was: | /* confdefs.h */ […] | | #include <stdio.h> | main () { | extern unsigned GC_version; | FILE *msg = fdopen(6,"w"); | unsigned major, minor, alpha; | major = GC_version >> 16; | minor = (GC_version >> 8) & 0xff; | alpha = GC_version & 0xff; | if (alpha == 0xff) fprintf(msg,"(version %d.%d found) ", major, minor); | else fprintf(msg,"(version %d.%d alpha %d found) ", major, minor, alpha); | return !( major > 7 || major == 7 && minor >= 2); } configure:9710: result: no, version at least 7.2 is required (gc will be built from downloaded sources) Looking at objdump -T /usr/lib/libgc.so, it seems that global variable might now be replaced by a function called GC_get_version. https://github.com/ivmai/bdwgc/commit/40a9f147990ed4668870100c897a8ded4754ba1f from 2009 introduces the function (since gc 7.2 alpha 4), while https://github.com/ivmai/bdwgc/commit/d6f15183b4064af11ce78cbf8a55edb04ad4a672 makes symbols like GC_version hidden by default (since 7.3 alpha 2). I guess the Macaulay configure script should be modified to use GC_get_version.
Reported upstream at https://github.com/Macaulay2/M2/issues/130
Thanks for reporting it upstream. I'm changing the title because it was misleading. We try to use as few bundled libraries as possible.
Created attachment 376302 [details, diff] Simple fix This patch is a kind of mixed approach fix: it uses preprocessor macros instead of data from the library, but it still runs the conftest program, instead of checking the condition at compile time. I'm not sure I'd code it that way for upstream, but for now, this fixes the issue and allows me to actually emerge the package.
OK, thanks. I'll wait a few days before I commit this, to see if Dan Grayson has any opinion on this. Macaulay2 has been very sensitive to switching boehm-gc versions in the past.
The problem with this patch is that it breaks gc detection with earlier versions. Is there any mechanism that works for both?
I wouldn't mind applying the suggested patch upstream, but it might be slightly better not to. The code in gc-7.4.0's file alloc.c that defines GC_version is this: #ifndef GC_NO_VERSION_VAR const unsigned GC_version = ((GC_VERSION_MAJOR << 16) | (GC_VERSION_MINOR << 8) | GC_VERSION_MICRO); #endif I've verified that Macaulay2's current gc presence test works with gc 7.4.0, as the symbol GC_version still gets defined. The string GC_NO_VERSION_VAR appears nowhere else, so the code is always compiled. Did the gentoo developers decide to define that macro variable when building gc? If so, why not revert that instead, for backward compatibility?
(In reply to Daniel R. Grayson from comment #6) > I wouldn't mind applying the suggested patch upstream, but it might be > slightly better not to. The code in gc-7.4.0's file alloc.c that defines > GC_version is this: > > #ifndef GC_NO_VERSION_VAR > const unsigned GC_version = ((GC_VERSION_MAJOR << 16) | > (GC_VERSION_MINOR << 8) | GC_VERSION_MICRO); > #endif > > I've verified that Macaulay2's current gc presence test works with gc 7.4.0, > as the symbol GC_version still gets defined. > > The string GC_NO_VERSION_VAR appears nowhere else, so the code is always > compiled. Did the gentoo developers decide to define that macro variable > when building gc? If so, why not revert that instead, for backward > compatibility? Hmm, I don't understand what's going on. As far as I can tell Gentoo's gc has not set GC_NO_VERSION_VAR and the header gc_version.h is present and included in gc.h. From looking at it, this part of the code is also unchanged between 7.2 which is detected by M2's configure on Gentoo and 7.4 which is not detected.
(In reply to Thomas Kahle from comment #7) > (In reply to Daniel R. Grayson from comment #6) > > I wouldn't mind applying the suggested patch upstream, but it might be > > slightly better not to. The code in gc-7.4.0's file alloc.c that defines > > GC_version is this: > > > > #ifndef GC_NO_VERSION_VAR > > const unsigned GC_version = ((GC_VERSION_MAJOR << 16) | > > (GC_VERSION_MINOR << 8) | GC_VERSION_MICRO); > > #endif > > > > I've verified that Macaulay2's current gc presence test works with gc 7.4.0, > > as the symbol GC_version still gets defined. > > > > The string GC_NO_VERSION_VAR appears nowhere else, so the code is always > > compiled. Did the gentoo developers decide to define that macro variable > > when building gc? If so, why not revert that instead, for backward > > compatibility? > > Hmm, I don't understand what's going on. As far as I can tell Gentoo's gc > has not set GC_NO_VERSION_VAR and the header gc_version.h is present and > included in gc.h. From looking at it, this part of the code is also > unchanged between 7.2 which is detected by M2's configure on Gentoo and 7.4 > which is not detected. Maybe one difference between your installation of gc and Gentoo's is --fvisibility-hidden when compiling gc ?
Good point: that visibility option kicks in for you but not for me: I build gc with enable-static and disable-shared. You probably do the opposite, and the configure script of gc makes the visibility conditional on those options. I'll patch upstream as suggested.
Alright, thanks.
Applied upstream at https://github.com/Macaulay2/M2/commit/eca1bacffcd8bb6f12a0f4e539e8cc87edfc5b9e
