Serializer_v5._pack_revision and _unpack_revision to become public?

[Denys Duchier]

John A Meinel <john at arbash-meinel.com> writes:

> Well David was saying that it is expensive to get revisions. But that
> should only be reading XML files. repository.get_revision() doesn't have
> to do anything but read XML.

He should really check if appropriate caching is done.  But, in any case, what
is expensive is the reading, not so much the XML.  It is possible to be
marginally faster by not using XML, but that's just the wrong thing to optimize
at present (unless things have recently changed dramatically).

> I realize that our overall performance is actually usually limited by
> O(n_lines) performance of weaves (especially for the inventory).
> And _extract tends to dominate times.

not _extract itself, but the computation of inclusions.  once the inclusions are
computed, extraction is quite fast (at least, that was so when I last
benchmarked these things).

Cheers,

--Denys