Add a -l --listen option to set the socket path or address:port
Fix a bug using thread safe functions from libfcgi

Now libtoolize copy missing files
Adds m4 macro directory for aclocal
etc.

This overflow was avoided using the volatile i

Declaring a counter as volatile to avoid GCC opti :/

The problem seems to appear when a worker fails to wait indicate it is idle
by waiting the WSTATE semaphore (maybe when pool handler waited the sem
to check pool's busyness and before the pool handler post it again...)

When running dummy scripts the sem_timedwait was unable to detect that the
pool was busy...

Defining a PYFCGI_STATS_SZ = 15 * 60s = 900

- send buffer formated by stats functions to clients
- close client socket if needed
- allow bind to reuse an address if nobody listen on it

The worker indicate each new request using semaphore

Implementing a dummy stream response loop sending PyFCGI version

Creating new files to handle IPC stuffs : include/ipc.h src/ipc.c

Now logging a python traceback

7 seems better than 5.

The responder_loop handle better child creation & deletion. A conf struct
member indicates the number of seconds the pool have to be idle before deleting
a worker.
In a same way, the pool has to be busy for at least 1s before a new
process can be added (the -f flag indicate if the counter has to be reseted
when a new process is created, allowing spawn burst).
The spawn() function call sigaction 2 times for SIGINT & SIGTERM in order
to catch both.

Adds a global flag for worker indicating their status (add checks for
each semaphore change)
The pool handler will no longer have to decrease the semaphore value when
a child exit.

Add a -t --timeout and a -f --fast-spawn option for further commit