On 5/14/2007 3:35 PM, Tom Lane wrote:
> Jan Wieck <JanWieck@Yahoo.com> writes:
>> The only problem with that is that there are code paths that set
>> ActiveSnapshot to palloc()'d memory that is released due to a
>> MemoryContextDelete() without resetting ActiveSnapshot to NULL.
>
> Only at the very end of a transaction (where ActiveSnapshot *is* reset
> to null, in FreeXactSnapshot); otherwise we'd have bugs unrelated to
> RevalidateCachedPlan. Eventually I would like to have reference-counted
> snapshots managed by a centralized module, as was discussed a month or
> two back; but right at the moment I don't think it's broken and I don't
> want to spend time on intermediate solutions.
Which means that the 8.3 fix for the reproducible backend crash, I
posted earlier, is to have SPI_cursor_open() save and restore
ActiveSnapshot while calling RevalidateCachedPlan(). I'll cross check
that this fixes this symptom and commit later today.
Jan
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#================================================== JanWieck@Yahoo.com #