I blogged quite a bit in the past about the real mess caused by the -fvisibility=hidden flag that KDE 3.4+ insisted to enable when finding some of the conditions needed to get it working.
After a lot of discussions with other devs, after looking at the logic, and after looking at the code, I decided that, while it’s a big size save option, it’s too unstable to get used at all, as both GCC is too untested and the change of a default parameter to linking is bad by itself.
The problem was originally blamed to Qt missing visibility support, then to pre-GCC4 patches by Gentoo and other distros, then by GCC 4.0.1 that was broken, then by another patch neede to 4.0.2 (and that added a lot of ICEs and crashes here and there), then by STL not supporting it yet…
Well today I was looking at KDE’s admin dir to fix the kde eclass a bit, and found this out:
------------------------------------------------------------------------ r480611 | mueller | 2005-11-15 19:38:50 +0100 (mar, 15 nov 2005) | 3 lines disable -fvisibility=hidden by default. Too many gcc bugs to workaround on popular linux distributions (that break LSB compliance)
compare this with KDE bug #109386. I said on 2005-09-27 to add a configure switch for that, as it was not safe. I continued reporting the problems, but seems like Dirk stopped looking at them in the mean time.
Now, it breaks LSB compliance, so it must be disabled, eh, it’s logical… when it breaks RedHat, is a true problem, when Gentoo devs tell that it’s unsafe and should be disabled by default it’s just shit, eh?
Oh well, I’m happy to be proved right about that, now I can remove the pissed off status caused by ALSA.
