On Mon, Feb 16, 2026 at 1:18 PM Nisha Moond <[email protected]> wrote: > > On Sat, Feb 14, 2026 at 3:45 AM Andres Freund <[email protected]> wrote: > > > > > > conflict.h includes utils/timestamp.h, even though it only needs the > > types. Using datatype/timestamp.h suffices (although it requires some fixups > > elsewhere). > > > > It works even without including "utils/timestamp.h" in conflict.h, as > removing this header still allows the build and full tests to pass. > This is likely because the required definitions are included > indirectly, as all files using conflict.h already include either > "access/commit_ts.h" (which includes "datatype/timestamp.h") or > "datatype/timestamp.h" itself. > > To avoid possible issues in the future, if conflict.h is used in a > file that does not include timestamp.h, it would be safer to include > "datatype/timestamp.h" directly, which is sufficient and requires no > fixups elsewhere. >
Yeah, that sounds better. > The attached patch includes the required change, and make check-world > passes cleanly for it. > Good. I have not tested it myself but if others don't see a problem with this, I can do the tests and push it. -- With Regards, Amit Kapila.
