OK, so I've reworked the patch and it doesn't break anything so do people want to see it or shall I just commit?
Basically we now cache the APR_ values and you can set the TCP_CORK/TCP_NOPUSH via the apr_setsocketopt call. apr_getsocketopt works fully as well (using the cached values...) david ----- Original Message ----- From: "Greg Stein" <[EMAIL PROTECTED]> To: "APR Development List" <dev@apr.apache.org> Sent: Tuesday, January 09, 2001 2:48 AM Subject: Re: [PATCH] track socket options... > On Mon, Jan 08, 2001 at 10:30:32PM -0000, David Reid wrote: > > Folks, > > > > This is a small patch that's very much a starting point. Basically > > following Deans suggestion we should be tracking what options we have set on > > a socket and using the cached information to avoid system calls. I've used > > an int and the system defined values as that way we can use the mask for non > > APR values, i.e. things like TCP_CORK. By caching the values we also get a > > working apr_getsocketopt which is useful. > > >From <asm/socket.h> on my Linux box: > > #define SO_DEBUG 1 > > >From <netinet/tcp.h>: > > #define TCP_NODELAY 0x01 > > > In other words... you have a collision. > > Told ya to use the APR values :-) > > (for things like TCP_CORK, just have an internal flag; or possibly go ahead > and place it into apr_network_io.h; I mean why not?!) > > (oh... and if you create internal flags: leave a comment in apr_network_io.h > about where the other flags are to prevent confusion / collision) > > The patch seems to go in the right direction. The big thing will be whether > APR has some internal uses of setsockopt that will need to tweak the flags. > (or just recode the internal one to use apr_setsockopt) > > Cheers, > -g > > -- > Greg Stein, http://www.lyra.org/ >
