On Tue, May 12, 2020 at 11:14:02AM +0000, JonY wrote: > >> I think option #3 is the simplest approach in the short term, though I > >> prefer option #2. > >> > > > > This patch has been accepted by MSYS2 [1], tested for bootstrapping on > > both i686 and x86_64. > > > > So here comes a 4th option: Disable `-Werror` if `--disable-werror` is > > specified to the top-level configure. > > > > EGREP'ing for `(format|__format__)\s*\(\s*(__printf__|printf)` in GCC > > source reveals ~30 matches (excluding testsuites). Probably they should > > all be fixed similarly, or we ignore such warnings for simplicity. > > libgomp doesn't build despite `--disable-werror` at top level, which > > seems a bug in this case. > > > > > > [1] > > https://github.com/msys2/MINGW-packages/blob/9501ee2afc8d01dc7d85383e4b22e91c30d93ca7/mingw-w64-gcc/0020-libgomp-Don-t-hard-code-MS-printf-attributes.patch > > > > > > Any thoughts on the libgomp printf attribute changes? gomp_fatal etc. call vfprintf under the hood though, does that one handle %llu on mingw? If not, using %llu for the PRI* macros looks wrong to me. Jakub
