Hi
While translating the v15 message catalog (yes, I'm quite late!), I
noticed that commit 1f39bce02154 introduced three copies of the
following message in hashagg_batch_read():
+ ereport(ERROR,
+ (errcode_for_file_access(),
+ errmsg("unexpected EOF for tape %d: requested %zu bytes, read %zu bytes",
+ tapenum, sizeof(uint32), nread)));
These messages should only arise when a hash spill file has gone
corrupt: as I understand, this cannot happen merely because of a full
disk, because that should fail during _write_ of the file, not read.
And no other user-caused causes should exist.
Therefore, I propose to turn these messages into errmsg_internal(), so
that we do not translate them. They could even be an elog() and ditch
the errcode, but I see no reason to go that far. Or we could make them
ERRCODE_DATA_CORRUPTED.
BTW, the aforementioned commit actually appeared in v13, and I
translated the message there at the time. The reason I noticed this now
is that the %d was changed to %p by commit c4649cce39a4, and it was that
change that triggered me now.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Small aircraft do not crash frequently ... usually only once!"
(ponder, http://thedailywtf.com/)
