Re: cvs: php-src / NEWS /ext/standard http_fopen_wrapper.c

> This morning I've started testing this new feature and I've found some bugs: > * it is segfaulting on redirects > I'll just bet I know why too.... The way the http:// wrapper is built, any redirects would be prone to stacking multiple filters on top of each other. There's two things to address here: (A) Why would the redirect indicate chunked encoding during a redirect? Better check for redirects and clear the autofilter flag in that case. (B) Why would multiply stacked filters segfault? Based on an earlier issue I had with zlib_filter I suspect this may relate to the handling of PSFS_ERR_FATAL (issued by the second itteration of http.chunked.decode because there's no chunksize to parse) in the filters api and isn't really a problem with chunked decoding per se. Your GDB and Valgrind output also seem to support that theory. > * the chunk decoding algorithm isn't working well. Just run a diff in thw > output of both HTTP/1.0 and 1.1. > That's interresting because the tests I ran got identical outputs both with and without chunked encode/decode. I didn't try google.com though, I'll give them a shot after I've addressed the larger issues. Sacha- I can appreciate that there is a limited potential need for PHP to support chunked encoding >right-now<, however this is one step in a larger process to support keep-alives and issuing multiple requests over a single http/https session. I'd like to work on the issues raised by Nuno: (A) being minor, and (B) being (seemingly) unrelated to chunked-encoding itself. If the release cycle starts to approach and we're no better off then obviously this can and should be torn out in the name of stability. -Sara