Only workdir is recreated, but e.g temp stays (with saved environment and other temp files potentially afecting the ebuild), distdir symlinks stay (thus SRC_URI change is not taken into account). It should perform a full equivalent of 'ebuild foo.ebuild clean' when .ebuild is touched. Partial cleanup is no good. As an possible enhancement, there could be a switch that would tell ebuild to ignore ebuild change completely and clean nothing (for small ebuild fixes that don't require full recompile etc).
Created attachment 103526 [details, diff] run dyn_clean instead of just `rm -rf $WORKDIR`, but respect FEATURES=keepwork This patch runs the full dyn_clean() function but allows the ebuild mtime to be ignored if you have FEATURES=keepwork enabled (possibly overridden via the environment). It passes a parameter to dyn_clean() that causes it to keep ${PORTAGE_BUILDDIR}/distdir intact so that the unpack phase can proceed normally.
It fails to remove stuff from image with "Permission denied" (but explicit ebuild foo clean works), maybe because the privileges are already dropped? My FEATURES=autoconfig ccache collision-protect cvs distlocks parallel-fetch sandbox sfperms strict stricter userfetch userpriv usersandbox
Also, it doesn't recreate tempdir (which brokes epatch at least) and homedir (which I doesn't know what it's for :)
(In reply to comment #3) > Also, it doesn't recreate tempdir (which brokes epatch at least) and homedir > (which I doesn't know what it's for :) > Afaik homedir is for when an ebuild writes to $HOME, which afaik is remapped during phases to homedir instead of portage's home.
(In reply to comment #2) > It fails to remove stuff from image with "Permission denied" (but explicit > ebuild foo clean works), maybe because the privileges are already dropped? I guess that code should really run on the python side, prior to the setup phase. I'll see about migrating it over there.
Created attachment 103603 [details, diff] clean ${T} in addition to ${WORKDIR}, but respect keepwork and keeptemp At the moment, for this particular case, it's not very easy to trigger a full clean phase on the python side for this without doing some significant refactoring or adding some ugly conditional logic. This new patch will at least clean up ${T} and allow you to edit the ebuild without forcing recreation of WORKDIR if you have FEATURES=keepwork. After looking at this some more, the whole thing seems a little too automatic to me. I'm inclined to have the unpack phase die instead of do this automatic cleanup, but I'm no sure how users would feel about that.
This is in svn r5244.
This has been released in 2.1.2_rc3-r1.
Still got some problems, see this: ebuild foo.ebuild compile touch foo.ebuild USE=doc ebuild foo.ebuild.compile The doc flag isn't respected. My guess is that it reads the $T/environment before cleaning $T?
(In reply to comment #9) > Still got some problems, see this: > > ebuild foo.ebuild compile > touch foo.ebuild > USE=doc ebuild foo.ebuild.compile > > The doc flag isn't respected. My guess is that it reads the $T/environment > before cleaning $T? > Yes. I think the best solution is to remove the automatic behavior completely and simply rely on the ${PORTAGE_BUILDDIR}/.unpacked like we do in most phases. If the user has modified the ebuild, I think it should be their responsibility to run the clean phase if that's what they want.
OK. I've got used to prefix every command with "clean" already :/
If you'll have to manually run clean anyway then there's nothing left to fix. I'm sorry that the existing automatic behavior doesn't meet your needs. Even though it's clearly broken as it is, I'd prefer not support such automatic behavior.
