On Fri, Oct 14, 2022 at 07:35:20PM +0000, Andrew Cooper wrote: > On 13/10/2022 14:05, Anthony Perard wrote: > > No need for $(AUTOSRCS), GNU make can generate them as needed when > > trying to build them as needed when trying to build the object. Also, > > those two AUTOSRCS don't need to be a prerequisite of "all". As for > > the "clean" target, those two files are already removed via "_*.c". > > > > We don't need $(AUTOINCS) either: > > - As for both _libxl_savm_msgs*.h headers, we are adding more > > selective dependencies so the headers will still be generated as > > needed. > > - "clean" rule already delete the _*.h files, so AUTOINCS aren't needed > > there. > > > > "libxl_internal_json.h" doesn't seems to have ever existed, so the > > dependency is removed. > > > > Add few prerequisite for "libxl_internal.h" so all headers that it > > depends on should be generated. And have $(SAVE_HELPER_OBJS) depends > > on "libxl_internal.h". > > > > Rework objects prerequisites, to have them dependents on either > > "libxl.h" or "libxl_internal.h". "libxl.h" is not normally included > > directly in the source code as "libxl_internal.h" is used instead. But > > we are adding "libxl.h" as prerequisite of "libxl_internal.h", so > > generated headers will still be generated as needed. > > > > "testidl.c" doesn't depends on "libxl.h" but "testidl.o" does. > > I'm afraid I don't follow here. How can this be true?
>From make point-of-view, in order to generate "testidl.c", we only need to execute "gentest.py" which takes "libxl_types.idl" as input. It doesn't even matter if "libxl.h" exist when generating "testidl.c" via `make testidl.c`. "libxl.h" is only used later when compiling "testidl.c" into "testidl.o". I can probably expand the commit message to better explain this. Thanks, -- Anthony PERARD
