* Charles Wilson wrote on Wed, Jan 21, 2009 at 10:10:08PM CET: > Ralf Wildenhues wrote: > > Part (1) is easy to review: it is obvious that regressions are very > > unlikely to be system-dependent. One does get the impression that it > > might just be more efficient to let libtool save the cwrapper text > > somewhere and the program just cat that. But still, this part is ok, > > please apply. > > This part pushed...
Thanks. > > Why is this patch not accompanied by a testsuite addition using > > -std=c89 -Werror on a program that creates a C wrapper? > > ...but without an additional test. Ralf, how should such a test be > structured? Do we need (like Darwin) a separate category of windows-ish > tests, that are skipped elsewhere, or what? I don't see a need to skip the test elsewhere. Here's what I'd do: transform $LIBTOOL to have CFLAGS and LTCFLAGS also contain -std=c89 -Werror. (The test would be even cleaner with a re-configured libtool, but let's not go overboard here.) With that, compile a library, and a program linked against it (so that, on w32, a wrapper is compiled). In order to avoid false failures due to non-GCC or so, you can also compile a trivial program and skip if the above flags cause an error. Cheers, Ralf
