This can't be the bug because it's in the input filter code, but the most recent commit to mod_deflate.c has some UB in it surrounding the case when readbytes is larger than what's available in ctx->proc_bb. In that case bkt is the sentinel for ctx->proc_bb and calling apr_brigade_split_ex looks wonky.
Anyhow, HTH. Index: mod_deflate.c =================================================================== --- mod_deflate.c (revision 910064) +++ mod_deflate.c (working copy) @@ -1011,11 +1011,13 @@ } if (!APR_BRIGADE_EMPTY(ctx->proc_bb)) { - /* May return APR_INCOMPLETE which is fine by us. */ - apr_brigade_partition(ctx->proc_bb, readbytes, &bkt); - - APR_BRIGADE_CONCAT(bb, ctx->proc_bb); - apr_brigade_split_ex(bb, bkt, ctx->proc_bb); + if (apr_brigade_partition(ctx->proc_bb, readbytes, &bkt) == APR_INCOMPLETE) { + APR_BRIGADE_CONCAT(bb, ctx->proc_bb); + } + else { + APR_BRIGADE_CONCAT(bb, ctx->proc_bb); + apr_brigade_split_ex(bb, bkt, ctx->proc_bb); + } } return APR_SUCCESS; ----- Original Message ---- > From: Eric Covener <cove...@gmail.com> > To: dev@httpd.apache.org > Sent: Sun, February 14, 2010 1:03:27 PM > Subject: Re: unsupported compression issue seen in 2.3.5-alpha > > On Sun, Feb 14, 2010 at 1:00 PM, Paul Querna wrote: > > curl -H 'Accept-Encoding: gzip;' -iL > > http://httpd.apache.org/docs/2.2/style/css/manual.css 2>/dev/null | > > head -12 > > > > What i've been told is, most of the time, this 'works' and you get > > compressed data. But sometimes, the same request returns uncompressed > > data, but with the same headers -- implying that mod_deflate was > > removed after adding the headers somehow -- this of course will cause > > browsers to try to decompress it, but since its already decomrpessed, > > it doesn't work. > > > > As far as I know, it has only been reported on 2.3.5 -- not in the > > earlier 2.3.3 (which the EU machine also ran before 2.3.5). If you > > search twitter for 'apache.org' you also see several more reports. > > FWIW, I'd never seen it before this saturday where I saw it for about > an hour consistently -- browser independent. C-E: gzip was set but > not actually compressed. > > > -- > Eric Covener > cove...@gmail.com