On 23/04/2020 15:48, Luboš Luňák wrote:
Just for completeness, there's one more option, and that's using Clang-cl instead of MSVC for everything. E.g. Mozilla does it, it is already possible to build LO that way (and I think Stephan does that regularly?). I'm not promoting this option though, as there are drawbacks to it too. Clang-cl currently builds noticeably more slowly then MSVC with enabled-by-default PCHs (for half a year I still haven't managed to get into Clang my PCH patches *sigh*), and I don't know how much testing Clang-cl-built LO has received. At least at this point it's probably better to just go with the gbuild hack I've created.
I routinely make (--enable-dbgutil etc.) builds with clang-cl from a locally-built LLVM/Clang trunk (i.e., not using whatever Visual Studio might offer) on Windows. For quite a while now, they consistently fail during `make check` in CppunitTest_sc_financial_functions_test and CppunitTest_sc_mathematical_functions_test, for reasons I have not yet explored. (My main motivation for those builds is to have loplugin also cover Windows-only code.)
_______________________________________________ LibreOffice mailing list LibreOffice@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/libreoffice
