Re: [GENERAL] Notify argument?

Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> The breakage will come when we lengthen NAMEDATALEN, which I plan to
> tackle for 7.3. We will need to re-order the NOTIFY structure and put
> the NAMEDATALEN string at the end of the struct so differing namedatalen
> backend/clients will work. If you want to break it, 7.3 would probably
> be the time to do it. :-) Users will need a recompile pre-7.3 to use
> notify for 7.3 and later anyway.

If we're going to change the structure anyway, let's fix it to be
independent of NAMEDATALEN. Instead of

char relname[NAMEDATALEN];
int be_pid;

let's do

char *relname;
int be_pid;

This should require no source-level changes in calling C code, thanks
to C's equivalence between pointers and arrays. We can preserve the
fact that freeing a PQnotifies result takes only one free() with a
little hacking to make the string be allocated in the same malloc call:

newNotify = (PGnotify *) malloc(sizeof(PGnotify) + strlen(str) + 1);
newNotify->relname = (char *) newNotify + sizeof(PGnotify);
strcpy(newNotify->relname, str);

Thus, with one line of extra ugliness inside the library, we solve the
problem permanently.

regards, tom lane