In [0] I had noticed that we have no automated verification that global
variables are declared in header files. (For global functions, we have
this through -Wmissing-prototypes.) As I mentioned there, I discovered
the Clang compiler option -Wmissing-variable-declarations, which does
exactly that. Clang has supported this for quite some time, and GCC 14,
which was released a few days ago, now also supports it. I went and
installed this option into the standard build flags and cleaned up the
warnings it found, which revealed a number of interesting things.
I think these checks are important. We have been trying to mark global
variables as PGDLLIMPORT consistently, but that only catches variables
declared in header files. Also, a threading project would surely
benefit from global variables (thread-local variables?) having
consistent declarations.
Attached are patches organized by sub-topic. The most dubious stuff is
in patches 0006 and 0007. A bunch of GUC-related variables are not in
header files but are pulled in via ad-hoc extern declarations. I can't
recognize an intentional scheme there, probably just done for
convenience or copied from previous practice. These should be organized
into appropriate header files.
[0]:
https://www.postgresql.org/message-id/c4ac402f-9d7b-45fa-bbc1-4a5bf0a9f206@eisentraut.org