Hello, gnustep-base fails to build docs: Making all for doc ReleaseNotes... ../Tools/obj/autogsdoc -Project ReleaseNotes -DocumentationDirectory ReleaseNotes -MakeDependencies ReleaseNotes/dependencies ReleaseNotes.gsdoc The file '/var/tmp/portage/gnustep-base/gnustep-base-1.22.1/temp/GNUstep.conf' is writable by someone other than its owner (permissions 0664). Ignoring it. make[1]: *** No rule to make target `ReleaseNotes/dependencies', needed by `internal-doc-all_'. Stop. make: *** [ReleaseNotes.all.doc.variables] Error 2 With USE="-doc" it builds fine.
Created attachment 291933 [details] build.log
Created attachment 291935 [details] environment
Created attachment 291937 [details] emerge --info
I have committed a fix to gnustep-base.eclass that fixes a sandbox violation (on tex fonts cache), but that's not the problem we have here... I suspect some of the user* FEATURES, as here without them the GNUstep.conf has correct permissions (644). I'll try to reproduce it
Portage team, maybe you can help: in gnustep-base.eclass, most phases begin with an environment check, which creates (if it does not already exists) ${T}/GNUstep.conf (to set temporary settings for sandboxing). Here for gnustep-base, it is created in src_configure(). However with FEATURES="userpriv usersandbox", at the end of src_configure the file has 664 permissions (the original is copied from /etc/GNUstep/GNUstep.conf and is 644). Apparently permissions for files in ${T} change at the end of the phase. But later in src_install(), the build system checks the GNUstep.conf and ignores it because of its incorrect permissions. So is there a way to keep the file permissions? I could do a chmod at the beginning of other phases, but that's more of a workaround.
(In reply to comment #5) > Apparently permissions for files in ${T} change at the end of the phase. But > later in src_install(), the build system checks the GNUstep.conf and ignores it > because of its incorrect permissions. > > So is there a way to keep the file permissions? I could do a chmod at the > beginning of other phases, but that's more of a workaround. If you instead use ${WORKDIR}/GNUstep.conf then the permissions will be preserved.
Created attachment 302175 [details, diff] gnustep-base.eclass patch
Indeed, that's a better way. Paweł, can you test with the attached patch to gnustep-base.eclass if this changes something to your sandbox problem? Thanks I will do some checks on the gnustep-* ebuilds before committing the change
(In reply to comment #8) > Paweł, can you test with the attached patch to gnustep-base.eclass if this > changes something to your sandbox problem? Thanks I can't reproduce bug #383665 now (without the patch). Not sure why. I didn't test the patch.
Ok, thanks. I committed the patch so this bug at least is fixed. Thanks everyone