Re: src/tools/pginclude considered harmful (was Re:

Bruce Momjian <bruce(at)momjian(dot)us> writes:
> FYI, 527 include were removed from non-header C files in this run. That
> is not something that can be easily done manually.

It's not so easily done automatically, either :-(. I'm not sure why
this go-round was so much more painful than the last, but it very
clearly exposed the deficiencies in your testing process. Mainly,
that you tested only one set of configure options on one platform.

I would say that minimum requirements for doing this again in future
are (1) test with and without --enable-cassert, and (2) test with and
without EXEC_BACKEND. We *know* that both those things change the
set of headers required. It might be necessary to actually test the
WIN32 port separately --- EXEC_BACKEND is close but not the same.

regards, tom lane