PG Bug reporting form <noreply@postgresql.org> writes:
> When testing PostgreSQL 12.2 on Windows built using MSVC (this potentially
> affects mingw too, though untested) on both Windows Server 2012 & 2016, we
> are having issues with a newly generated log file on a fresh PostgreSQL
> cluster when PostgreSQL is started with pg_ctl. If the log file was created
> with a prior version of PostgreSQL, or if you manually create the log file
> and grant the executing user the appropriate file access permissions,
> everything works fine.
> In other words, the issue occurs when passing a file name to the `-l` flag
> where the log file does not already exist.
So ... why aren't we seeing that in the buildfarm? The TAP tests
certainly do that, and I think the core regression tests do too.
> We think this was introduced in 0da33c76. One approach for a fix would be to
> add a condition so the added code only executes if the file already exists,
> and if the file does not exist it executes the "snprintf" command like
> before.
Without an explanation for why it's not failing everywhere, I don't
have a lot of faith in that being an improvement.
regards, tom lane