Re: Exceptional md.c paths for recovery and zero_damaged_pages

Andres Freund <andres(at)anarazel(dot)de> writes:
> We have a fair number of special paths in md.c that are specific to
> recovery. E.g. in mdreadv() we do:
> ...
> As far as I can tell, nearly all - including the above - InRecovery paths in
> md.c are basically unreachable. And have been for quite a while.

> XLogReadBufferExtended() takes care to
> a) Create the fork if it doesn't yet exist.
> b) Not to read from beyond EOF. If EOF is found, we extend the relation to be
> large enough.
> Which afaict should suffice to prevent needing the above?

I haven't checked the git history, but I suspect this logic is later
than the md.c code you mention, and may well render it obsolete.

> The InRecovery paths for _mdfd_getseg seem to originate in 2004's 303e46ea932
> and the zero-beyond-eof seems to be from 2007's ef07221997e - although it was
> just *restricted* to InRecovery in that commit.

We definitely needed 303e46ea932 at the time, but that doesn't mean
we still do. Possibly ef07221997e was just copying the earlier logic.

regards, tom lane