dev-libs/icu-51.1 has subslot 50.1
after updating from 50.1.2 all packages depend on icu (over sqlite[icu] too) will be broken.
so, subslot must be 51.1
/sorry my bad english
Sorry about that, its fixed. Thanks for the report.
>so, subslot must be 51.1
Your suggestion is wrong:
Nothing links to libicuuc.so.51.1 or libicuuc.so.51.2 in my systems, all packages link to libicuuc.so.51. When i updated icu from 51.1 to 51.2 emerge recompiled a lot packages without any good reason.
libicuuc.so.51.1 was not preserverd, 'emerge -vp @preserved-rebuild' was empty when emerge installed icu and was rebuilding packages.
Wasting time for rebuilding libreoffice is not fun, please fix SLOT in icu ebuilds.
add to make.conf:
As far as I can remember, icu consumers reacted rather sensitive to icu upgrades in the past. Even if revdep-rebuild did not rebuild (i.e. soname did not change), random crashes and other irregularities occured until the consumers were rebuilt. So from me, this is for now a wontfix.
Then add comment to ebuild near SLOT= about this problem, else it is far from obvious.
(In reply to Andreas K. Hüttel from comment #4)
ICU provides 5 public libraries. In the past, there happened sonameless changes of ABI of 1 of these libraries (libicule), which is used by almost no reverse dependency. (libicule is used by app-office/libreoffice.)
(Subslot dictionaries will be proper solution. ICU will need 8 items in its subslot dictionary.)
(In reply to Nikoli from comment #5)
> Then add comment to ebuild near SLOT= about this problem, else it is far
> from obvious.
Yes, it is recommended to do that.
(In reply to Julian Ospald (hasufell) from comment #7)
> (In reply to Nikoli from comment #5)
> > Then add comment to ebuild near SLOT= about this problem, else it is far
> > from obvious.
> Yes, it is recommended to do that.
Done, with a reference to this bug here.
IMHO we can revisit this in a few releases and relax the condition if we think upstream behaves more sanely. Or if we have the subslot dicts.