Re: Fix some error handling for read() and errno

On Thu, Jun 21, 2018 at 2:32 AM, Michael Paquier <michael(at)paquier(dot)xyz> wrote:
> Sure. I have looked at that and I am finishing with the updated version
> attached.
>
> I removed the portion about slru in the code, but I would like to add at
> least a mention about it in the commit message. The simplifications are
> also applied for the control file messages you changed as well in
> cfb758b. Also in ReadControlFile(), we use "could not open control
> file", so for consistency this should be replaced with a more simple
> message. I noticed as well that the 2PC file handling loses errno a
> couple of times, and that we had better keep the messages also
> consistent if we do the simplification move... relmapper.c also gains
> simplifications.
>
> All the incorrect errno handling I found should be back-patched in my
> opinion as we did so in the past with 2a11b188 for example. I am not
> really willing to bother about the error message improvements on
> anything else than HEAD (not v11, but v12), but... Opinions about all
> that?

I think this should be split into two patches. Fixing failure to save
and restore errno is a different issue from cleaning up short write
messages. I agree that the former should be back-patched and the
latter not.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company