Hi, as seen in bug #139693, built_with_use will return 0 (True) if the package's USE *file* doesn't exist. However, a package can't possibly be built with any USE *flag* if it's not even installed (i.e. built). In built_with_use() in eutils.eclass, the function should distinguish between the package being installed, not installed, package.provided or injected. It's not obvious how to do that, though. If has_version were aware of package.provided and injected packages, then built_with_use() could contain near its start: if ! has_version $1 ; then # Package is not installed return 1 fi
Created attachment 93098 [details, diff] portageq.patch Fixes has_version to look in package.provided.
Created attachment 93099 [details, diff] eutils.eclass.patch Fixes built_with_use to use the fixed has_version.
if you're using built_with_use on a package that isnt installed, then your ebuild logic is flawed in a larger sense perhaps the correct fix would be to just call 'die'
(In reply to comment #3) > if you're using built_with_use on a package that isnt installed, then your > ebuild logic is flawed in a larger sense Nope, it is Portage's logic that is flawed (and fixed by these patches). Having to work around Portage's flaws in ebuilds, rather than fixing Portage's commands so that they do not produce the *opposite* result of what anyone would reasonably expect, is no way to program. There are two bugs which these patches fix: * has_version ignores package.provided. * built_with_use is wrong, as mentioned above. Should I split this into 2 bugs rather than 1?
two different issues file a new bug about the has_version as that is a portage issue as for built_with_use, my previous comment still stands
Created attachment 93613 [details] irc.log Here's the discussion on IRC, for reference. Note that the sentence "i can come up with a circumstance going the opposite direction where yours breaks but mine works" on line 94 turns out to be false.
*** Bug 163435 has been marked as a duplicate of this bug. ***
