2319 2012-04-17 17:32:08 +0000 gthumb: do_install hangs on highly parallel builds 2012-08-30 12:22:25 +0000 1 1 7 Build System, Metadata & Runtime OE-Core multimedia 1.2 x86 Multiple VERIFIED FIXED (P1)(Patch Review) Medium normal 1.3 M3 1 dvhart ross.burton meta.mr.watcher meta.watcher mihai.lindner richard.purdie ross.burton sgw song.liu --- --- --- oldest_to_newest 20553 0 dvhart 2012-04-17 17:32:08 +0000 OE Build Configuration: BB_VERSION = "1.15.1" TARGET_ARCH = "i586" TARGET_OS = "linux" MACHINE = "qemux86" DISTRO = "poky" DISTRO_VERSION = "1.1+snapshot-20120417" TUNE_FEATURES = "m32 i586" TARGET_FPU = "" meta meta-yocto = "dvhart/tiny:c1922f73308b4c2ef021c55adb3cdc0f25cb45bc" gthumb do_install will hang with make running at 100% CPU. At RP's suggestion I set PARALLEL_MAKEINST="" in the gthumb recipe and it successfully installed. This may or may not be related to Bug 2245 (gthumb fails to install configration files correctly). 22801 1 ross.burton 2012-07-04 15:46:34 +0000 I'm pretty sure that this was due to the gconftool race. At least I can't replicate the failure after fixing gconf.bbclass with -j8 when I remove PARALLEL_MAKEINST. 22803 2 ross.burton 2012-07-04 17:08:56 +0000 Patch sent to oe-core. 22804 3 richard.purdie 2012-07-04 19:13:22 +0000 Sadly I've just seen gthumb hang in compile so there is a problem with this recipe. This was with Ross' gconf changes applied. ps ax shows: 30934 pts/2 S+ 0:00 /bin/sh /media/build1/poky/scripts/bitbake gthumb -c install 30952 pts/2 S+ 0:01 python /media/build1/poky/bitbake/bin//bitbake gthumb -c install 30953 pts/2 Sl+ 0:07 python /media/build1/poky/bitbake/bin//bitbake gthumb -c install 31048 pts/2 S 0:00 python /media/build1/poky/bitbake/bin//bitbake gthumb -c install 31049 pts/2 S 0:00 /bin/sh -e /media/build1/poky/build/tmp/work/armv7a-vfp-neon-poky-linux-gnueabi/gth 31050 pts/2 S 0:00 make -j 24 31053 pts/2 S 0:00 make all-recursive 31054 pts/2 S 0:00 /bin/bash -c fail= failcom='exit 1'; \?for f in x $MAKEFLAGS; do \? case $f in \? 31185 pts/2 S 0:00 /bin/bash -c fail= failcom='exit 1'; \?for f in x $MAKEFLAGS; do \? case $f in \? 31186 pts/2 S 0:00 make all 31187 pts/2 S 0:00 /bin/bash -c fail= failcom='exit 1'; \?for f in x $MAKEFLAGS; do \? case $f in \? 31322 pts/2 S 0:00 /bin/bash -c fail= failcom='exit 1'; \?for f in x $MAKEFLAGS; do \? case $f in \? 31323 pts/2 S 0:00 make all 31331 pts/2 S 0:00 make all-recursive 31332 pts/2 S 0:00 /bin/bash -c fail= failcom='exit 1'; \?for f in x $MAKEFLAGS; do \? case $f in \? 31345 pts/2 R 3:11 make all-am 31364 pts/1 S 0:00 bash 31429 pts/1 R+ 0:00 ps ax and the build log is pretty unhelpful. I'll try and figure this out further. 22805 4 ross.burton 2012-07-04 20:13:04 +0000 Hmpf. I'll try with -j24 tomorrow. I wonder what in the makefile is tripping up and spinning, from memory there wasn't anything complicated in it. 22813 5 richard.purdie 2012-07-04 21:37:55 +0000 I'm pretty sure there is some bug in the glib-mkenum usage in gthumb's makefiles. Removing all the glib-mkenum usage makes things work. With it left as is, it wants to rexecute these targets multiple times per build and bombs out with anything other than "make". Even make -j 2 fails. It seems to be the fact they're listed in BUILD_SOURCES which means "make all" triggers them first and this interacts badly somehow with make all-recursive. It seems these things are hard to get right and I noticed: https://bugzilla.gnome.org/show_bug.cgi?id=654395 https://mail.gnome.org/archives/networkmanager-list/2012-February/msg00070.html 22814 6 richard.purdie 2012-07-04 21:57:33 +0000 To confirm, the problem is in sections like: enum-types.h: $(HEADER_FILES) $(GLIB_MKENUMS) $(AM_V_GEN)( $(GLIB_MKENUMS) \ --fhead "#ifndef ENUM_TYPES_H\n#define ENUM_TYPES_H\n\n#include <glib-object.h>\n\nG_BEGIN_DECLS\n" \ --fprod "/* enumerations from \"@filename@\" */\n" \ --vhead "GType @enum_name@_get_type (void);\n#define GTH_TYPE_@ENUMSHORT@ (@enum_name@_get_type())\n" \ --ftail "G_END_DECLS\n\n#endif /* ENUM_TYPES_H */" \ $^> xgen-$(@F) \ && (cmp -s xgen-$(@F) enum-types.h || cp xgen-$(@F) enum-types.h ) \ && rm -f xgen-$(@F) ) where adding a "touch enum-types.h" to the end works around the problem. make is hanging waiting for the file to get updated but never sees the timestamp change it wants as far as I can tell. 22815 7 richard.purdie 2012-07-04 22:24:33 +0000 and finally a reproducer note. Try "touch xxxx/glib-mkenum" where xxxx is the path within the sysroot (pretend glib was reinstalled), then force a recompile. 23666 8 ross.burton 2012-07-31 15:57:18 +0000 I totally have a patch for this. 23675 9 ross.burton 2012-07-31 17:14:02 +0000 Patch sent to oe-core. 23842 10 ross.burton 2012-08-06 21:21:42 +0000 This was merged earlier today. (From OE-Core rev: dea6a0eb2bf06b3dc47e276a046d8dd828e47228) 24663 11 mihai.lindner 2012-08-30 12:22:25 +0000 Cannot reproduce with current master and `bitbake ghtumb`. BB_VERSION = "1.15.3" TARGET_ARCH = "i586" TARGET_OS = "linux" MACHINE = "qemux86" DISTRO = "poky" DISTRO_VERSION = "1.2+snapshot-20120830" TUNE_FEATURES = "m32 i586" TARGET_FPU = "" meta meta-yocto = "master:bb0aab79e488a99cb1b43f72f0e8ddbe39d97519"