Re: FATAL: bogus data in lock file "postmaster.pid": ""

On Tue, Aug 28, 2012 at 04:25:36PM -0400, Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > Updated patch attached which just reports the file as empty. I assume
> > we don't want the extra text output for pg_ctl like we do for the
> > backend.
>
> The backend side of this looks mostly sane to me (but drop the \n,
> messages are not supposed to contain those). But the feof test proposed

Removed. I thought we needed to add \n so that strings >80 would wrap
properly. How do we handle this?

> for pg_ctl is no good: consider a file containing just, say, "-".
> fscanf would eat the "-", then hit eof, and this would complain the file
> is empty. Possibly checking for ftell(pidf) == 0 would do, though I'm
> not sure whether it's portable to assume fscanf would eat a non-numeric
> character before complaining.

ftell() seems to work fine when combined with feof(), so I used that in
the attached patch. ftell() alone remains at zero if the file contains
"A", so feof() is also needed.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ It's impossible for everything to be true. +