El divendres, 30 de setembre de 2016, a les 11:05:19 CEST, René J.V. Bertin va escriure: > Hello, > > I think I've tried to raise some awareness about this already in the past. > The issue has come up again among KDE/Mac users, and this time I decided to > file a ticket on BKO. Finding the right category wasn't trivial, so I > picked what seemed the best compromise and relay it here: > > https://bugs.kde.org/show_bug.cgi?id=369554 > > I don't really know how invasive it would be to address this issue properly, > or rather, how to fix this without being overly invasive. Looking at > KPeople I see > > /opt/local/include/KF5/KPeople/KPeople/Widgets/Actions > /opt/local/include/KF5/KPeople/kpeople/widgets/actions.h > > which means that on a case-insensitive-but-preserving FS (HFS, NTFS) you can > end up with 4 different situations: > > /opt/local/include/KF5/KPeople/kpeople/widgets/{Actions,actions.h} > /opt/local/include/KF5/KPeople/KPeople/widgets/{Actions,actions.h} > /opt/local/include/KF5/KPeople/kpeople/Widgets/{Actions,actions.h} > /opt/local/include/KF5/KPeople/KPeople/Widgets/{Actions,actions.h} > > and that's presuming that the case preservations works reliably. Transfer > such a tree via a FAT32 thumbdrive and all bets are off.
Why would you do that? Honestly I think you're trying to fix a corner case that noone is really facing. Cheers, Albert > Evidently that's > less likely to happen with a headerfile directory (and just shouldn't be > done) but I'm concerned that the underlying implicit assumption about the > FS has been made elsewhere too. And as I note in the ticket: Mac OS app > bundles should support being transferred to and from case-insensitive > volumes and still be usable on a case-sensitive volume (idem for KF5 > framework bundles if they ever see the light). > > The most straightforward fix would be to use only CamelCase OR lowercase > header directories, i.e. one of > /opt/local/include/KF5/KPeople/kpeople/widgets/{Actions,actions.h} > /opt/local/include/KF5/KPeople/KPeople/Widgets/{Actions,actions.h} > > This is also what Qt5 does, and one can presume they have seen a lot more > testing of this approach on many more platforms. NB: one could create the > appropriate symlinks when installing to a case-sensitive FS, but evidently > that makes it impossible to copy the result to a case-insensitive FS. > > I realise that doing this is an invasive change that requires patching lots > of dependent code, and even if those patches could be generated > automatically the patched code would then no longer build against older > versions of the frameworks. Still, it'd be an almost perfect way to force > people to look for other places where the assumption about FS > case-sensitivity could lead to problems. > > R