On Friday 2014-10-03 13:19, Holger Hans Peter Freyther wrote:
On Fri, Oct 03, 2014 at 12:57:03PM +0200, Jan Engelhardt wrote:
When --disable-talloc was given to configure, it would already link libosmocore.so to libtalloc. Make it use pkg-config to discover talloc's flags and location, and when the system talloc is used, make sure to use its API header file too.
include/osmocom/core/talloc.h.in | 6 ++
Have you considered patching libosmocore.pc.in as well to list talloc as "required" there? This way users of libosmocore do not need to check for talloc themselves?
I can do that.
AC_ARG_ENABLE(talloc, [AS_HELP_STRING( [--disable-talloc],
[Disable building talloc memory allocator]
[Disable using bundled talloc memory allocator]AM_CONDITIONAL(ENABLE_TALLOC, [test x"$enable_talloc" = x"yes"]) +AS_IF([test x"$enable_talloc" = xno],
I think the wording does not reflect the two different usecases we have here.
- Firmware builds with no talloc
That may be the use case, but it is not what was implemented. If you build 0.7.0 [336915aa0a] with --disable-talloc, `ldd src/.libs/libosmocore.so` will show libtalloc.so.2. Which kind of goes against this use case. Therefore, "disable-talloc" really just meant "disable-system-talloc".
diff --git a/include/osmocom/core/talloc_int.h b/include/osmocom/core/talloc_int.h new file mode 100644 index 0000000..dde102b --- /dev/null +++ b/include/osmocom/core/talloc_int.h
Do you know the '-M' option for send-email/format-patch? It helps with reviewing simple renames.
Been there, tried that. And when I used it, comes someone who wants a non-M submission because they are using some traditional utility or something. Not here, but I remember it occurring on netfilter-devel.
diff --git a/tests/msgfile/msgfile_test.c b/tests/msgfile/msgfile_test.c index a919694..2684b6a 100644 --- a/tests/msgfile/msgfile_test.c +++ b/tests/msgfile/msgfile_test.c @@ -20,6 +20,7 @@ */
#include <osmocom/core/msgfile.h> +#include <osmocom/core/talloc.h>
Is this hunk needed?
Yes, because system talloc has a
#define talloc_free(ctx) _talloc_free(ctx, __location__)
and otherwise, one gets "implicit definition of talloc_free" and "undefined reference to `talloc_free'".
If you address the pkg-config issue I am happy to inclue your patch. In terms of options.. maybe we can just remove --disable-talloc instead of having sort of a tri-state?
What's the third option?