On Tue, Oct 15, 2013 at 10:48:19AM -0400, Ryan Johnson wrote:
Hi all,
While trying to build python3 for cygwin, I kept encountering the
following error message:
./Modules/signalmodule.c: In function ?fill_siginfo?:
./Modules/signalmodule.c:745:60: error: ?siginfo_t? has no member named
?si_band?
PyStructSequence_SET_ITEM(result, 6, PyLong_FromLong(si->si_band));
^
Include/tupleobject.h:62:75: note: in definition of macro
?PyTuple_SET_ITEM?
#define PyTuple_SET_ITEM(op, i, v) (((PyTupleObject *)(op))->ob_item[i]
= v)
^
./Modules/signalmodule.c:745:5: note: in expansion of macro
?PyStructSequence_SET_ITEM?
PyStructSequence_SET_ITEM(result, 6, PyLong_FromLong(si->si_band));
As far as I can tell, siginfo_t::si_band is mandated by POSIX.1-2001,
and required for proper handling of SIGPOLL. The latter seems to
correspond to async I/O with poll(2). I'm pretty sure cygwin doesn't
support async I/O, but shouldn't the struct member at least exist, to
avoid breaking code that assumes its existence? The alternative is to
patch python3 locally so its os.sigwaitinfo function no longer touches
si_band, or to file a bug upstream so that the module's configury tests
for its existence before using it.
Thoughts?
Sure. I question the utility of lying in a structure about the
availability of an unimplemented feature. If something is specifically
expecting the structure member to exist it seems like it would be
expecting it to do something.