On Sun, Dec 05, 2004 at 10:35:11AM -0700, Paul Querna wrote:
Jeff Trawick wrote:
On Sun, 05 Dec 2004 00:01:05 -0700, Paul Querna <[EMAIL PROTECTED]> 1) write_full() has the wrong semantics for apr_file_writev();
apr_file_writev() and apr_file_write() are not supposed to block
Thanks, I didn't think about the issue of write_full() blocking. r109892 uses apr_file_write(), and adds the check for an under-write as you originally suggested.
...and back to the beginning, did you read the original thread on this topic?
No, I did not remember or was not aware of the 'original thread'.
I was looking at use iovecs in mod_log_config, and in IRC Justin mentioned that the !HAVE_WRITEV implementation sucks. That started me down the path of trying to fix it.
I assume you are talking about this thread in October? http://marc.theaimsgroup.com/?t=109693523200002&r=1&w=2
The fact that the caller must cope with short writes is exactly
why the original implementation was valid and correct. All subsequent
versions have been variously buggy - the type of tbytes is still wrong. I think this should just be reverted back to the original version.
joe
