Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> wrote:
> On 30.05.2011 17:10, Kevin Grittner wrote:
>> Heikki Linnakangas wrote:
>>> * XXX: for an anomaly to occur, T2 must've committed
>>> * before W. Couldn't we easily check that here, or does
>>> * the fact that W committed already somehow imply it?
>>
>> The flags and macros should probably be renamed to make it more
>> obvious that this is covered. The flag which SxactHasConflictOut
>> is based on is only set when the conflict out is to a transaction
>> which committed ahead of the flagged transaction. Regarding the
>> other test -- since we have two transactions (Tin and Tout) which
>> have not been summarized, and transactions are summarized in
>> order of commit, SxactHasSummaryConflictOut implies that the
>> transaction with the flag set has not committed before the
>> transaction to which it has a rw-conflict out.
>
> Ah, got it.
>
>> The problem is coming up with a more accurate name which isn't
>> really long. The best I can think of is
>> SxactHasConflictOutToEarlierCommit, which is a little long. If
>> nobody has a better idea, I guess it's better to have a long name
>> than a misleading one. Do you think we need to rename
>> SxactHasSummaryConflictOut or just add a comment on that use that
>> a summarized transaction will always be committed ahead of any
>> transactions which are not summarized?
>
> Dunno. I guess a comment would do. Can you write a separate patch
> for that, please?
Attached is a comments-only patch for this, along with one
correction to the comments you added and a couple other typos.
I'll submit a patch for the README-SSI file once I find a reference
I like to a paper describing what Dan's proof uses as a premise --
that the transaction on the rw-conflict *out* side of the pivot must
not only be the first of the three transactions in the dangerous
structure to commit, but the first in the entire cycle of which the
dangerous structure is a part. With that premise as a given, the
proof is short, clear, and unassailable; but I think we need a cite
to use that argument convincingly.
-Kevin