It might be done by tomorrow. I think that this would not have been a problem for any linux tests. I read once that linux has a high IOV_MAX -- but of the two seemingly identical debian boxes I have here, one seems to have 16 and the other 1024. But I suspect that writev/readv ignore this value anyway, and process arbitrary numbers of iovecs, because the linux man pages don't mention it. And Windows doesn't have an equivalent restriction either.
Jim
William A. Rowe, Jr. wrote:
At 05:32 PM 4/23/2003, Jim Carlson wrote:
I'm considering writing a patch to the unix version of apr_socket_sendv(), which would check IOV_MAX and issue several calls to sendv() if necessary (while respecting the blocking, nonblocking, timeout setting in the same way as apr_socket_send()). I have to write this code anyway, but it seems to belong in APR. Any objections?
None whatsoever. If it is ready tomorrow, I will even review it in the afternoon, since we already other Win32 socket issues that I'm vetting.
I'm only surprised that we haven't seen this in Apache with mod_include abuse tests (which create very long chains of iovecs, or at least they used to.)
Perhaps this is the root of the various "only the first <--! include>'ed document is processed by ssi" bug reports? And why most people don't see them?
Bill
