REGRESSION in 0.17rc1, doing commit of files with unicode names.

[Martin Pool]

On 6/18/07, Ian Clatworthy <ian.clatworthy at internode.on.net> wrote:
> Martin Pool wrote:
>
> > Yes, just reverting it like that is probably the way to go for 0.17.
> > Ian did you find it was much faster to go direct to stderr?
>
> Martin,
>
> It was faster going just to stderr, particularly on the initial commit.
> There was a long debate on buffering of messages, if we should bother
> and if we did, where to do it. That all got backed out because going to
> stderr removed the need for it (performance wise) and no-one agreed with
> me that it was correct semantically.
>
> I'm going to look at commit again as part of 0.18 so I'll revisit the
> whole logging & progress bar thing then. In fact, hopefully later today!

I have a long-delayed branch that changes us from using the Python
logging library, which seems measurably slow (though not dominant for
most operations.)

  https://code.launchpad.net/~mbp/bzr/logging

The way we test logging and rely on it while testing other things
makes this slightly complicated.

> I like Aaron's idea of stopping note() logging to the file at all. Is
> that an acceptable approach to have product wide?

Do you mean to .bzr.log, or are you talking about the effect of --quiet?

-- 
Martin