Re: Minimal logical decoding on standbys

Hi,

On 2023-Apr-03, Andres Freund wrote:

> Hm? That's what the _'s do. We build strings in parts in other places too.

No, what _() does is mark each piece for translation separately. But a
translation cannot be done on string pieces, and later have all the
pieces appended together to form a full sentence. Let me show the
"!terminating" case as example and grab some translations for it from
src/backend/po/de.po:

"invalidating" -> "... wird ungültig gemacht" (?)

(if logical) " obsolete replication" -> " obsolete Replikation"

" slot \"%s\" because it conflicts with recovery" -> " Slot \"%s\", weil sie in Konflikt mit Wiederherstellung steht"

If you just concatenate all the translated phrases together, the
resulting string will make no sense; keep in mind the "obsolete
replication" part may or not may not be there. And there's no way to
make that work: even if you found an ordering of the English parts that
allows you to translate each piece separately and have it make sense for
German, the same won't work for Spanish or Japanese.

You have to give the translator a complete phrase and let them turn into
a complete translated phrases. Building from parts doesn't work. We're
very good at avoiding string building; we have a couple of cases, but
they are *very* minor.

string 1 "invalidating slot \"%s\" because it conflicts with recovery"

string 2 "invalidating obsolete replication slot \"%s\" because it conflicts with recovery"

(I'm not clear on why did Bertrand omitted the word "replication" in the
case where the slot is not logical)

I think the errdetail() are okay, it's the errmsg() bits that are bogus.

And yes, well caught on having to use errmsg_internal and
errdetail_internal() to avoid double translation.

Cheers

--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/