On Mon, Jul 29, 2013 at 11:44 AM, Reid Kleckner <[email protected]> wrote: > On Mon, Jul 29, 2013 at 11:33 AM, Hans Wennborg <[email protected]> wrote: >> >> On Mon, Jul 29, 2013 at 11:01 AM, NAKAMURA Takumi <[email protected]> >> wrote: >> > I suppose that system's python should be /usr/bin/python in Cygwin's >> > world. >> > I didn't know Python/win32 could run Lit for cygwin tests. >> > >> > What is a motivation for you to take *external* python on cygwin? >> >> I build Clang outside Cygwin, in cmd.exe, because I want to build in a >> "pure win32" environment. But I fail at running the tests in cmd.exe, >> so I use Cygwin to run the tests. >> >> If I try to use Cygwin's /usr/bin/python to run the tests, it fails >> because of some Windows path. Probably because I didn't build in >> Cygwin. >> >> I guess my configuration is confusing to lit.cfg, because I have >> platform.system() = "Windows" and lit.getBashPath() = >> "C:\src\cygwin\bin\bash.EXE". But it's still not MSYS :) > > > This configuration is pretty handy because it's way faster to spawn > processes from native win32 python than from cygwin bash or cygwin python. > > I think the odd duck that we should be checking for here is MSys, so using > the MSYSTEM env var sounds good to me.
I got paranoid and started worrying about the possibility of MSYS bash ending up on the path, even if the test suite is started outside an MSYS shell. In that case, looking for $MSYSTEM wouldn't work. Maybe we should just ask the bash binary which version it has? I'm attaching a patch that does that, please take a look. I realize this isn't super important, as the tests are still being run by the cygwin buildbots, but it would be nice if this configuration were supported too. Thanks, Hans
tighten-shell-preserves-root2.diff
Description: Binary data
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
