Created attachment 401838 [details, diff] Partial patch (no nls) against bash ebuild with USE=static support Original discussion in https://bugs.gentoo.org/show_bug.cgi?id=7714 There were some counter-suggestions like providing separate package for static bash or using sash. Separate package sounds like maintenance burden and sash is not a bash, hence the bug to possibly revisit the idea. Attached patch requires updating - virtual/libintl needs static-libs USE flag. (all providers, namely gettext - already have it)
Some rationale: Downgrade of sys-libs/readline breaks bash already during the emerge leaving the system without working primary shell. Dependency constraints cannot prevent the downgrade when app-shells/bash is downgraded with sys-libs/readline at the same time. Hence the idea to introduce static bash version, perhaps even enable the flag by default.
(In reply to Maciej Mrozowski from comment #0) please read bug 7714 in full first. specifically, comment #3. (In reply to Maciej Mrozowski from comment #1) SONAME changes aren't available (older ones are SLOT-ed), and even then, current readline-6.x packages shouldn't be a big deal ...
Well, to me, if anything, comment #3 from bug 7714 is rather after providing private dynamic bash version (with private shared library deps) solely for portage purpose, not for forbidding static system bash. > SONAME changes aren't available (older ones are SLOT-ed), and even then, current readline-6.x packages shouldn't be a big deal ... But they are. This is what prompted me to open the bug, I switched to latest ~arch bash (4.2->4.3, needs readline 6.2->6.3) (to test bug 542796), then switched back (both bash and readline - switching readline was obviousy a mistake). Now, I'm not saying bash downgrade is common/standard use-case...
(In reply to Maciej Mrozowski from comment #3) yes, we could in theory build a dynamic bash just for ebuild usage. but that isn't worth the trade-off imo. i don't think the bash downgrade was your problem. more likely the readline downgrade which used the same SONAME but had fewer ABI exports.
Indeed that was the case.
