I may be missing something here, ap_content_length_filter looks broken.
Currently it implements an unlimited size buffer, by trying to morph
every indeterminate length bucket into the heap. It has the standard
"read till it blocks then flush" logic, but this isn't a defence against
If the (e.g) CGI script is fast enough that read()s never block it just
keeps on sucking up HEAP buckets, as the simple repro case in the bug
1. am I being stupid here?
2. otherwise, is there a good defence of why that filter should buffer
to try to compute a C-L - exactly how much it should buffer & why?!