Re: refactor backend type lists

On 2025-Jul-28, Euler Taveira wrote:

> On Tue, Jul 15, 2025, at 3:30 PM, Álvaro Herrera wrote:

> > Second, in discussion [2] leading to commit 18d67a8d7d30 (Nov 2024) it
> > was agreed to add support for translating backend type descriptions.
> > I'm not really sure that this is useful.  It would be, if set_ps_display
> > and pg_stat_activity used translated names, so that you could match what
> > log messages say with what the process lists show.  But I think we've
> > historically not translated those.  We have a few translatable strings
> > as the argument of HandleChildCrash() in postmaster.c, but those are
> > using names that are yet a third source of strings; I'm not a fan of
> > this.  (For staters, if a translation decided to use non-ascii chars for
> > process names, would that work okay in set_ps_display?  I bet it
> > wouldn't, because that's using strlen()).  So I would propose to rewind
> > a bit here, and remove translation from all those places so that the
> > output is consistent (== usable) between log messages and ps/pg_stat_activity.
> 
> I'm not sure if it is a good idea to have translated backend description. The
> init_ps_display() output is certainly used to obtain information (PID?) from a
> process list. There is also the option %b from log_line_prefix that is used to
> filter log messages per backend type. The same applies to backend_type column
> from pg_stat_activity view.

I should have let you know that I spent some time on this today as well
to avoid duplicating efforts.  Here are my patches, incorporating your
fixup -- I hadn't looked at your 0004 yet, so I wrote it differently,
passing the BackendType enum directly to LogChildExit (as well as
HandleChildCrash), so it is the former function that does the call to
GetBackendTypeDesc() to obtain the string.  I think this is better
because the untranslated part of the sentence is enclosed in quotes,
which I think is better.  However it meant I had to add a bit of a hack
to cope with the background worker separate bgw_type string.

So I came up with things as attached, incorporating your 0003 fixup.
What do you think?


This is likely not final, because the lines for background workers look
a bit inconsistent:

2025-07-28 23:10:02.316 CEST worker_spi dynamic[1876557] FATAL:  terminating background worker "worker_spi dynamic" due to administrator command
2025-07-28 23:10:02.317 CEST postmaster[1876543] LOG:  "background worker" process of type "logical replication launcher" (PID 1876552) exited with exit code 1


(I, for one, would be VERY HAPPY to not have to translate the phrase
"background worker".  There's just no reasonable way.)

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/