On Mon, Jan 17, 2022 at 5:14 PM Paul Eggert <egg...@cs.ucla.edu> wrote: > > On 1/17/22 10:37, Reuben Thomas wrote: > > > "extern-inline" copies a stanza of code into > > config.h. That code includes a test of the preprocessor macro > > "__header_inline". That macro is defined in a system header on macOS. > > When config.h is #included, no other header has been included. > > Therefore, the symbol will never have been defined at the point where > > it is tested in config.h. Therefore, the test will always fail. > > Thanks for explaining more. That part of Gnulib was introduced in 2013: > > https://lists.gnu.org/r/bug-gnulib/2013-11/msg00045.html > > and as I vaguely recall was in response to this macports ticket: > > https://trac.macports.org/ticket/41033 > > Apparently until now nobody tested that part of Gnulib to make sure that > it omits the workaround on newer Apple platforms where the underlying > bug has been fixed. (The underlying bug was that <ctype.h> defined > static functions to implement isalpha etc. on OS X 10.8 and earlier in > C, but the C standard does not allow this and using static functions > broke some GNU code.) > > Although the workaround still functions on current macOS, it's better to > use proper inline functions when available so I installed the attached > patch into Gnulib. Please give it a try. > > I don't know whether the macOS world has moved on and nobody cares about > older Apple releases with the ctype.h bug. If so, we could simplify Gnulib.
I think the version of OSX to test is OS X 10.5. I've got a PowerMac running it, if interested. Send over your SSH public key for access. Jeff