Description:
------------
first time loading the reproduction script works fine. subsequent reloads will first cause php.exe to crash (The instruction at "0x100108b4" referenced memory at "0x00000000". The memory could not be "read".), then once I hit OK on that alert, I'll get a "Server Error, unable to connect to fastcgi server." error in the browser. 

We've found that this error text is in the FastCGI source. It looks like maybe php crashes, then fastcgi attempts to access a php thread that has gone bye-bye, and throws that error. So there may be 2 issues: what's causing the original php crash, and then the fastcgi problem.

Here's how to repro:

- environment: win2k, IIS, FastCGI, PHP 4.3.2/.3 (I repro'd w/ both)
- set up a test script, such as a phpinfo script
- configure the repro script to hit your host, port, path_to_test_script and load it, then hit refresh (sometimes, I'll have to restart w3svc to be able to repro the bug)

notes:
- this does not occur if using fsockopen, only with pfsockopen
- i could not reproduce this with windows XP, only windows 2000 (both server and professional)


Reproduce code:
---------------
http://dietrich.ganx4.com/phpbug_iis.phps

Expected result:
----------------
it should output the URL it's attempting to connect to.

Actual result:
--------------
first time loading the reproduction script works fine. subsequent reloads will first cause php.exe to crash (The instruction at "0x100108b4" referenced memory at "0x00000000". The memory could not be "read".), then once I hit OK on that alert, I'll get a "Server Error, unable to connect to fastcgi server." error in the browser.