--continue-on-failure breaks elog-demo-hook distributed with paludis

[ddanier]

As the hook stated in its comments it should be put into the (un)install_fail-directory. If done so it is called when one of the packages in the list fails, which means during the build-process. Most of the time this erases the elog-temp-file and with it the information the elog-hook should provide _after_ paludis is finished. So log-entries get lost.

I don't know if putting the script into (un)install_all_post would be enough in all cases (--continue-on-failure if-fetch-only?), but if this hooks are called afterwards (even if an package failed) case I don't see any reason to keep the script in (un)install_fail. Perhaps the comments in the script can be updated to avoid confusing new users? If (un)install_all_post is not called when one out of many packages fails it probably should be (the hook-name does not say anything about success of the (un)install anyway). Or some hook could be introduced to handle this special case.

Tested with "--continue-on-failure if-independent", "if-satisfied" is untested, but should behave similar I think.

I first posted here this here (for reference):
 http://lists.pioto.org/pipermail/paludis-user/2007-November/000112.html

[anonymous]

Milestone Before 0.26.0 deleted

[ColdWind]

*_all_post are not executed when a package fails.

AFAIK the only way to make this work is executing *_all_post even when a package fails or creating a new hook which is executed on both cases (after *_all_post or *_fail).

[ciaranm]

r4502 provides new phases to deal with this.