POST request to fastcgi FPM with chunked encoding does not properly return result

Hello Xinchen,

since I've seen you did most of this work around fastcgi during this
transition, I was wondering if you maybe have any idea what the problem
could be?

thanks,

• Markus

PS: sorry for top-posting, but though it was more appropriate as there's
no context-specific question on my part yet.

Hello,

I wrote up a detailed question at SO but I had a hunch it could be a
real bug (
http://stackoverflow.com/questions/34440220/post-request-to-php7-with-chunked-encoding-does-not-properly-return-result
).

Basically, a POST request to php-fpm with "Transfer-Encoding: chunked"
and post data (could be even empty) does not return anything:

$ curl -XPOST http://localhost/ -H "Transfer-Encoding: chunked" -d ''
curl: (18) transfer closed with outstanding read data remaining

The PHP script could be anything returning a small amount of data, i.e.
below 4kb. A simple echo "world" would suffice.

The last commit I was able to compile which works was this one:
https://github.com/php/php-src/commit/16265a59ac6bd433bfb636e4e44da1ad57cdcda9

After that, for a few commits, the tree for me was in a state where I
couldn't compile it.

The first commit I was able to compile again, but which already exhibits
the problem, was
https://github.com/php/php-src/commit/86de98cabada88f4667839794c176ea37648498b
.

These are the commits in between I couldn't compile:

$ git bisect skip
There are only 'skip'ped commits left to test.
The first bad commit could be any of:
ba5ecf355fe792a5a2a8e6582d5e081d02b16fbf
e383cb4493031a7cd952cfcaed3297e583149c07
fef18f4bea1980a59a9283c2197bd090aaf500cb
18cf4e0a8a574034f60f4d123407c173e57e54ec
We cannot bisect more!

The diff is quite involving due files being moved around.

I'm not very experienced in this area so I'm not sure if the curl call
should work; however it does work fine in PHP 5.6. I can work around
this in my application by avoiding "Transfer-Encoding: chunked" but I
actually want to understand the issue better here, maybe it's even a bug?

thanks,

• Markus