On Mon, 5 Nov 2018 at 16:11, Burton, Ross <[email protected]> wrote: > On Mon, 5 Nov 2018 at 11:43, Burton, Ross <[email protected]> wrote: > > Mining the list archive shows that we keep on getting sqlite upgrades > > which fail like this. Maxin reported that it was due to the > > full-text-search code enabling itself and failing to build, so one > > solution is to explicitly turn it off. > > Did some digging. > > FTS5 is enabled by default, this needs log() from libm. If enabled, > AC_SEARCH_LIBS is used to find log() and libm.so, and this bonus > dependency that is needed if statically linking is recorded in > sqlite3.pc. Pseudo then builds, statically links to pseudo, but also > hard-codes the link line instead of using pkgconfig so this -lm isn't > added, resulting in a link failure on upgrade. > > So that explains why the autobuilder explodes when this upgrade > happens, but doesn't explain how it builds for everyone who submits > the upgrade...
This patch I can confirm makes pseudo build: -PACKAGECONFIG ?= "" +PACKAGECONFIG ?= "fts5" PACKAGECONFIG_class-native = "" PACKAGECONFIG[editline] = "--enable-editline,--disable-editline,libedit" PACKAGECONFIG[readline] = "--enable-readline,--disable-readline,readline ncurses" +PACKAGECONFIG[fts5] = "--enable-fts5,--disable-fts5" But that's a little ugly. The alternative is more invasive and would involve teaching pseudo to use pkgconfig to find sqlite3, but I'm not convinced we want to add pkg-config-native to the depends on pseudo, so I'm wondering if the third way is to just make pseudo always link to libm.so. Ross -- _______________________________________________ Openembedded-core mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-core
