Re: svn commit: rev 6110 - trunk/subversion/svnserve

On Mon, 2003-06-02 at 13:45, Philip Martin wrote:
> Do you consider it worse than in the original patch I posted?

I'm not sure.

> Part of the problem (the only tricky bit in my view) is that the fork
> case follows standard practice for pool reuse, while the thread case
> needs a separate pool for each thread. Changing the fork case to
> behave like the thread case would simplify things. I image any
> performance loss due to creating/destroying rather than
> clearing/reusing a pool is negligible in comparison to fork.

Having two different kinds of pool management for the same while loop is
definitely a big part of the reason I don't like the current structure
of the main loop. I'm fine with changing the fork case to create and
destroy a pool each time, and am willing to review how the code looks
after that change.

On the other hand, the amount of code shared between the fork case and
the thread case seems very small: a call to apr_accept() and an error
check. So I'm not sure why it wouldn't be better to have two
independent loops.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Jun 2 20:08:35 2003