lubot | [telegram] <RikMills> @mitya57 if qt6-base need lto disabling, would you say the rest of the stack does as well? | 09:05 |
---|---|---|
lubot | [telegram] <RikMills> Or do we need to work through the stack one by one? | 09:06 |
lubot | [telegram] <mitya57> I don't know. In Qt 5 times build flags from qtbase applied everywhere, but Qt 6 has a different build system so this may have changed. | 09:38 |
lubot | [telegram] <RikMills> Trying qt base anyway, then will see what happens I guess. I note most of qt5 seems to be in the lto-disabled-list package | 09:43 |
lubot | [telegram] <mitya57> Ah, that will explain it. | 09:45 |
lubot | [telegram] <mitya57> Adding to that list would be much better than patching everything. | 09:45 |
lubot | [telegram] <RikMills> Indeed | 09:46 |
lubot | [telegram] <RikMills> qt6-base build with LTO off, but then fails on ppc64el on symbols | 10:57 |
lubot | [telegram] <RikMills> sigh.... | 10:57 |
lubot | [telegram] <mitya57> I can mark those symbols as optional in Debian. | 10:58 |
lubot | [telegram] <RikMills> That would be good. I'll retry builds in a clean PPA 1st, as armhf in the current one actually built with LTO, so I want to start there from scratch (re @mitya57: I can mark those symbols as optional in Debian.) | 11:00 |
lubot | [telegram] <RikMills> Then I guess I can follow this: https://www.processon.com/view/link/61f50f161efad479c0785a08 | 11:00 |
lubot | [telegram] <RikMills> @mitya57 tempting to just shove all of Qt is the lto-disable-list and be done with it, but I guess it should all be checked propperly? | 11:02 |
lubot | [telegram] <mitya57> If Qt 5 was there, then it makes sense. | 11:04 |
lubot | [telegram] <RikMills> Ok, I will compromise then. Build qt6-base, then try the next tier in the Qt6 stack. If most of that also fails with LTO, then I'll stop and propose the whole stack goes into the lto-disabled-list. Reasonable? (re @mitya57: If Qt 5 was there, then it makes sense.) | 11:07 |
lubot | [telegram] <mitya57> https://matterbridge.lubuntu.me/e0896d46/file_3938_tgs.webp | 11:08 |
lubot | [telegram] <RikMills> Nice sticker! | 11:08 |
lubot | [telegram] <RikMills> so far only qt6-qtbase seems to have an LTO issue | 15:54 |
lubot | [telegram] <RikMills> symbols are an issue though | 15:54 |
lubot | [telegram] <RikMills> wondering if some of those are LTO related | 15:58 |
fvogt | How do you enable LTO? If you just pass -flto, that won't work | 16:01 |
fvogt | You'd need to use DCMAKE_INTERPROCEDURAL_OPTIMIZATION:BOOL=ON and let Qt/CMake handle it itself | 16:02 |
fvogt | Then it disables LTO on the version tagging object. | 16:02 |
RikMills | fvogt: LTO is turned on via a patched dpkg in Ubuntu | 16:06 |
fvogt | I suspected something like that | 16:06 |
RikMills | https://launchpad.net/ubuntu/+source/dpkg/1.20.7.1ubuntu2 | 16:07 |
RikMills | https://launchpad.net/ubuntu/+source/dpkg/1.20.7.1ubuntu4 | 16:08 |
fvogt | You'll have to opt out of that and go the cmake way | 16:08 |
RikMills | if I end up adding the qt6 packages to the excluded list it will never get enabled | 16:09 |
lubot | [telegram] <RikMills> yes, disabling LTO seems to make most symbols issues go away. probably some slight delta to go to debian for ppc64el symbols | 16:38 |
fvogt | You don't have to disable LTO, just disable the dpkg injection of -flto and set CMAKE_INTERPROCEDURAL_OPTIMIZATION instead | 16:40 |
RikMills | not going to carry such a delta to debian. adding all qt6 to the lto-disabled list until debian enabled LTO by default seems the best thing here | 16:41 |
lubot | [telegram] <RikMills> https://launchpad.net/bugs/1961324 | 17:17 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!