On Mon, 31 Aug 2020 18:52:35 -0500 Ken Moffat via blfs-dev wrote: > On Mon, Aug 31, 2020 at 10:32:05PM +0100, Ken Moffat via blfs-dev > wrote: > > On Mon, Aug 31, 2020 at 07:17:30PM +0100, Ken Moffat via blfs-dev > > wrote: > > > On Mon, Aug 31, 2020 at 08:04:37AM -0500, Marty Jack via > > > blfs-dev wrote: > > > > > > > > > > > > On Mon, 31 Aug 2020 01:44:03 -0500 Ken Moffat via blfs-dev > > > > wrote: > > > > > On Mon, Aug 31, 2020 at 06:48:47AM +0100, Ken Moffat via > > > > > blfs-dev wrote: > > > > > > > > > > Changing the title ecause this is a separate issue. > > > > > > > > > > > > Oh - tried updating libreoffice on another machine: with > > > > > > the skia code it defaults to using clang (and my CFLAGS, > > > > > > CXXFLAGS have -fstack-clash-protection which clang-10 does > > > > > > not understand). > > > > > > > > > > > > ĸen > > > > > > > > > > Nope, it still uses clang, which eventually fails: > > > > > > > > > > clang-10: error: unknown argument: > > > > > '-fstack-clash-protection' clang-10: error: unsupported > > > > > argument '8' to option 'flto=' > > > > > > > > > > I'm guessing clang generated flto=8 (an 8-thread machine) > > > > > and got confused by the previous unsupproted switch. > > > > > > > > > > https://www.phoronix.com/scan.php?page=news_item&px=LibreOffice-7.0-Prefers-Clang > > > > > > > > > > Oddly, Arch (libreoffice-fresh) do not seem to list clang or > > > > > even llvm as deps. > > > > > > > > > > Fedora are using --disable-skia which might work around > > > > > this. I'll try that for my own builds, which are a bt less > > > > > than the book (e,g, no java) but I think we'll need to make > > > > > clang required in the book. > > > > > > > [...] > > > > > > > > I build libreoffice 7 with gcc and without clang, which I do > > > > not have installed. I also do not have skia installed. It > > > > generates > > > > > > > > checking whether to build Skia... yes checking for clang... no > > > > configure: WARNING: Clang compiler not found. > > > > > > > > Takes about an hour on my i5-8600 for make build-nocheck. > > > > Clearly they may change this in future but this suggests that > > > > clang need not be required by the book. > > > > > > Hmm. Technically it is 'Recommended'. All my systems do have > > > clang and I don'y have a log from the failed build, all I know > > > for certain is that it tried to use clang when building skia, > > > and failed because clang does not (in current release) support > > > my flags. > > > > > > I'm guessing that your log can be interpreted as should I build > > > skia : yes (default) do you have clang : no ok, cannot build > > > skia, issue a warning. > > > > > > > I've just done a fresh build (omitting e.g. java, openldap, > > gssapi, krb5 which I have not installed) and using make V=1. > > Unfortunately, that no-longer gives a verbose build so I'm none > > the wiser about what actually got compiled. > > > > If anyone knows how to get libreoffice-7 (or late 6) to show the > > commands it is running, please speak up! > > > > Meanwhile, I've just started a build after hiding all the clang* > > files in /usr/bin. Not sure it will show anything. > > Unfortunately, there seems to be a reluctance in some build > > systems (e.g. typical python modules) to provide real verbose > > output. > > > Completed, the only apparent difference in the logs is timings/dates > in downloads, and some items get built at a different place which is > probably just random variation on a machine where I've got a desktop > with various open terms and browsers. > > In the output from autogen, it definitely could not now find clang, > so in that sense my build without clang matched yours: > > --- LO-logs-with-clang/mylog-autogen 2020-08-31 19:51:55.335992586 +0100 > +++ LO-logs-no-clang/mylog-autogen 2020-09-01 00:26:15.004670799 +0100 > @@ -386,16 +386,9 @@ > checking cpp/poppler-version.h presence... yes > checking for cpp/poppler-version.h... yes > checking whether to build Skia... yes > -checking for clang... clang > -checking for clang++... clang++ > -checking whether clang++ can compile SSE2 intrinsics... yes > -checking whether clang++ can compile SSSE3 intrinsics... yes > -checking whether clang++ can compile SSE4.1 intrinsics... yes > -checking whether clang++ can compile SSE4.2 intrinsics... yes > -checking whether clang++ can compile AVX intrinsics... yes > -checking whether clang++ can compile AVX2 intrinsics... yes > -checking whether clang++ can compile F16C intrinsics... yes > -checking whether clang++ can compile FMA intrinsics... yes > +checking for clang... no > +checking for clang++... no > +configure: WARNING: Clang compiler not found. > checking which gpgmepp to use... external > checking gpgme++/gpgmepp_version.h usability... yes > checking gpgme++/gpgmepp_version.h presence... yes > > > Based on the report from phoronix, and my own failure because clang > rejected one of my flags, I think that clang is intended to be used > (i.e. recommended), but I'm now less clear on the consequences of > using --disable-skia. Certainly it prevents clang being used if it > is present. > > ĸen > -- > I could not live without Champagne. In victory I deserve it, in > defeat I need it. -- Churchill > -- > http://lists.linuxfromscratch.org/listinfo/blfs-dev > FAQ: http://www.linuxfromscratch.org/blfs/faq.html > Unsubscribe: See the above information page
This reference https://skia.org/user/build#compilers claims that both gcc and clang will work but that clang is preferred because performance might be "dramatically worse" with other compilers. Your experiment seems to confirm that skia is built in both clang-present and clang-absent cases. I note that Arch doesn't have a standalone skia package. It looks tough to build standalone. I am successfully using the "qt5" (non KDE) VCL plug but I also build the "gtk3" that I use only when I have to produce PDF because at the moment "qt5" produces corrupt PDF and they say it is a lot of work to fix. I have no performance issues that I notice. Personally I don't expect to have clang on my system until it can be used to build everything and the LFS toolchain converts to it. Mainly because I don't like to have two of anything unless it is unavoidable. -- http://lists.linuxfromscratch.org/listinfo/blfs-dev FAQ: http://www.linuxfromscratch.org/blfs/faq.html Unsubscribe: See the above information page
