Michael Olson <[EMAIL PROTECTED]> writes:

> And now we have two calls to external programs rather than one, and we
> are still parsing the output of the entire log message.  And unless I'm
> missing something, which is possible since I'm not very familiar with
> ewoc, we will have to be continually seeking (either by regexp or by
> text property) into the output buffer in order to place new commit
> information

Likely to be true, yes. I don't know ewoc enough either.

> -- my method would instead jump to a single marker without search
> and start inserting output. I don't think this new option would gain
> us anything.

I'm not worried about the result of your solution, but about the
implementation effort. The advantage of this solution is that it keeps
the current process-management. But it's not necessarily better, that
was just an idea.

>> Indeed, we have something like this for baz, it keeps a cache (both on
>> disk and in memory) of the revision information.
>
> And I really hated the need for that "feature", because it caused a lot
> of ssh instances to be run for remotely-hosted archives when getting
> revision information for the first time.  Some hosting sites would
> simply refuse to connect after a certain number of ssh connections
> within a certain timeframe, which causes cascading error messages for up
> to 10 minutes.

The problem was different for baz/tla. We needed some ancestry
information that wasn't provided on the command-line, so we _had_ to
call a command (cat-log IIRC) for each revision. And the problem was
that we couldn't group the calls to cat-log (like "tla cat-log rev1
rev2 rev3 ...").

Indeed, I had implemented a fix for that (baz log --complete-log), but
the Bazaar team decided to throw my code away instead of ever
releasing baz 1.5 :-(.

-- 
Matthieu

_______________________________________________
Dvc-dev mailing list
[email protected]
https://mail.gna.org/listinfo/dvc-dev

Reply via email to