On Tue, 2 Jul 2002, Justin Erenkrantz wrote:
> Fix forthcoming (we're discussing alternatives on IRC now). -- justin
The results of said conversation:
Index: protocol.c
===================================================================
RCS file: /home/cvs/httpd-2.0/server/protocol.c,v
retrieving revision 1.107
diff -u -d -r1.107 protocol.c
--- protocol.c 26 Jun 2002 19:45:07 -0000 1.107
+++ protocol.c 2 Jul 2002 23:44:54 -0000
@@ -1199,12 +1199,14 @@
split = NULL;
flush = 0;
- APR_BRIGADE_FOREACH(e, b) {
+ e = APR_BRIGADE_FIRST(b);
+ while (e != APR_BRIGADE_SENTINEL(b)) {
const char *ignored;
apr_size_t len;
len = 0;
if (APR_BUCKET_IS_EOS(e)) {
eos = 1;
+ break;
}
else if (APR_BUCKET_IS_FLUSH(e)) {
if (partial_send_okay) {
@@ -1241,6 +1243,7 @@
flush = 1;
break;
}
+ continue;
}
else if (rv != APR_EOF) {
ap_log_rerror(APLOG_MARK, APLOG_ERR, rv, r,
@@ -1255,6 +1258,7 @@
ctx->curr_len += len;
r->bytes_sent += len;
+ e = APR_BUCKET_NEXT(e);
}
if (split) {
The key part to notice here is that when we get APR_EAGAIN from
apr_bucket_read(), we were failing to ever read from that bucket again
because of the APR_BRIGADE_FOREACH.
--Cliff
