Hi hackers,
> See about line 300 in src/backend/Makefile. In any case, it's
> easy to show by experiment that it does.
`make maintainer-clean` did the trick, thanks. I suggest modifying meson.build
accordingly:
-run make distclean in the source tree.
+run `make maintainer-clean` in the source tree.
> I think this is timescale's issue. Why are you defining / undefining
> PGDLLEXPORT?
That's a great question.
As I understand some time ago the developers had a problem with a collision of
exported symbols on *nix platforms [1] and chose to solve it by re-defining
PGDLLEXPORT to __attribute__((visibility ("default"))) for GCC and CLang.
I agree that this is a questionable approach. Redefining a macro provided
by Postgres doesn't strike me as a good idea. I tried to remove this
re-definition, but it didn't go well [2]. So apparently it should be addressed
somehow differently.
> Part of the patch series is to use visibility attributes, and your #if
> TS_EMPTY(PGDLLEXPORT) thing can't handle that.
Out of curiosity, how come a patchset that adds an alternative build system
changes the visibility attributes? I would guess they should be the same
for both Autotools and Meson. Is it necessary in order to make Meson work?
If not, maybe it should be a separate patch.
[1]: https://github.com/timescale/timescaledb/commit/027b7b29420a742d7615c70d9f19b2b99c488c2c
[2]: https://github.com/timescale/timescaledb/pull/4413
--
Best regards,
Aleksander Alekseev