From: "Bill Stoddard" <[EMAIL PROTECTED]>
Sent: Friday, July 27, 2001 10:57 AM
> In the mod_asis handler, we call:
>
> apr_file_open()
> ap_send_fd()
> apr_file_close()
>
> The problem is that the brigade created by ap_send_fd is not be flushed to the
>network
> before the apr_file_close (because we've not hit the min bytes to write threshold
>and one
> of the filters is buffering). I see a couple of solutions:
>
> 1. Not call apr_file_close() if we call ap_send_fd(). The bucket cleanup code will
>handle
> closing the file.
>
> 2. Chase the call to ap_send_fd() with a call to ap_rflush() and call apr_file_close
>as is
> happening now.
>
> 3. Variation on 2)... Pass an additional option to ap_send_fd to specify whether to
>flush
> or not. And call apr_file_close() as is happening now.
>
> I am leaning toward 3 for this specific case. This way, the handler can tell
>ap_send_fd
> to flush and be assured the bytes have been sent to the network before it returns.
>
> Thoughts?
None of the above? Have ap_send_fd play the same game as a bucket, duping and
refcounting
the file to a safe bucket?
