Hm, this should have been fixed already, see bug #69033. Anyway I cannot reproduce this with the built-in web server. Is there something special one needs to care about? Which server?

Thanks.

I was using an Apache 2.4 server.

I assume that the built-in web server would not be sufficient because it can only process a single request at a time without deadlocking. That is to say, I tried testing with the built-in server, but it hangs when I attempt to cURL on the same port with callee.php.

It should be noted that the test I had can be misleading, as a failed curl_exec could also output false (easily remedied by outputting something additional on callee.php).

It's also worth noting that I additionally tested this on 5.5.25 and got the same result. The patch in bug #69033 was included in 5.5.22.

@jake, sapi/apache utilizes thread safety, sapi/cli and sapi/fcgi not necessarily. But thead safety is probably be the exact difference.

That your CLI hangs must be some config issue, the built-in server shouldn't fail on such simple case. The key point of #69033 is that ENV isn't cleared within the same process. I'll test this with Apache then.

Thanks.

Here we go. putenv()/getenv() rely on C runtime and are process wide. The list of variables set by putenv() is maintained internally, those variables are then cleared on request shutdown.

With a thread safe SAPI, every request is served concurrently within the same process. Thus, the environment of the first request isn't cleared yet while the second request starts. But both are served within the same process, both are running simultaneously, so they share the environment. 

Thanks.