Neil Conway <neilc(at)samurai(dot)com> writes:
> AFAIK, ReadBuffer() will elog on error, so callers can assume that the
> buffer it returns is valid. The vast majority of ReadBuffer() call sites
> make this assumption, but some went to the trouble of checking that the
> returned buffer was valid and elog'ing if it was not. I've removed the
> error checking from the latter since it is dead code.
Agreed. I get the impression that at one time it was not so, but
certainly for the last many years there's been no need to test.
> I thought about adding an assertion (or even a precautionary
> elog(ERROR)) to ReadBuffer to verify that the returned buffer is indeed
> valid, but I didn't end up doing it. Feel free to raise your hand if you
> think this is a good idea.
Nah; considering that the return statements invoke
BufferDescriptorGetBuffer, you'll probably get a core dump anyway
if there's something wrong ;-)
A related issue in the same general area is that the smgr code is
currently implemented to elog on error, but its API still reflects
an assumption that it will return a failure indication. Changing
the API is a larger change than I want to see during late beta,
but it's a cleanup that would be reasonable to undertake during
a future development cycle, if you're interested.
regards, tom lane
In response to
pgsql-patches by date
|Next:||From: Tom Lane||Date: 2004-11-13 23:39:53|
|Subject: Re: Give the TODO list a little more verbose explanation |
|Previous:||From: Andrew Dunstan||Date: 2004-11-13 14:31:34|
|Subject: Re: Win32 signals & sockets|