svn commit: samba r7025 - in branches/SAMBA_3_0/source: lib tdb

[Jeremy Allison]

On Fri, May 27, 2005 at 04:07:56PM -0400, derrell at samba.org wrote:
> 
> tridge had specifically recommended against this because of the added overhead
> of read_socket_with_timeout() on each usage.  Thus I had written it to use
> read_socket_data() initially and only call read_socket_with_timeout() if
> EAGAIN occurred.  Since we have no way to "unread" the partial packet consumed
> by read_socket_data(), and you don't want to change the semantics of
> read_socket_data() to return a partial packet, this slower method may be the
> best we can do.
> 
> Do you want to check it in, shall I?

I just did :-). It's not slower in the server case as we call receive_smb and
friends with zero timeout, and only with a non-zero timeout in the case we're
processing an oplock break (which doesn't happen very often).

What did tridge say about the overhead in what cases and why didn't he say
it on the mailing lists ? :-). It was on bloody IRC, wasn't it.... Own up !

When will people realise that a discussion on IRC is a self-selecting club
of fans, and won't reach all the people affected by the decisions.

Jeremy.