Re: [HACKERS] interface libpq Makefile.in patch
От | Tom Lane |
---|---|
Тема | Re: [HACKERS] interface libpq Makefile.in patch |
Дата | |
Msg-id | 22143.918776542@sss.pgh.pa.us обсуждение исходный текст |
Ответ на | Re: [HACKERS] interface libpq Makefile.in patch (Tom Lane <tgl@sss.pgh.pa.us>) |
Список | pgsql-hackers |
>> The following patch is needed to compile the current development tree >> with perl support. The addition of MemoryContextAlloc, >> MemoryContextFree, and MemoryContexRealloc in >> ./src/include/utils/palloc.h require these additional header files. > Something wrong here ... palloc should not be visible outside the > backend. I have fixed the immediate symptom of the problem by making the Perl5 module not require libpq-int.h, which is better programming practice anyway. However, it is true that including libpq-int.h now requires access to backend include files that are not currently being installed into /usr/local/pgsql/include. Thus, compiling an outside application that uses libpq-int.h will presently fail. I had intended all along to someday stop exporting libpq-int.h, but I didn't really want to break code dependent on it this soon :-(. In any case, I think that the very same problem will occur for backend extension code (SPI) compiled outside the Postgres source tree --- the real problem is that "postgres.h" can't be included from the install tree anymore. I think we have two reasonable alternatives: (1) Install a bunch more backend-internals header files, along the lines of Ryan's proposed patch. Evidently we needinclude/utils/mcxt.hinclude/nodes/memnodes.hinclude/nodes/nodes.hinclude/lib/fstack.hinclude/utils/memutils.h and possibly other stuff. (2) Try to clean up the palloc macros so that they don't need quite as many random include files to be available. (Jan? Any chance of reducing the tonnage a little?) BTW, it'd really be a good idea to stop using libpq's makefile as the place where backend header files are installed... regards, tom lane
В списке pgsql-hackers по дате отправления: