Originally we decided that the output of VCC should be 100%
stand-alone and therefore contain no #includes at all.  This
was hoped to avoid unspecified trouble with C-compilers at runtime.

But C99 is old enough to drink now, so we move forward.

The script in tools/include_wash.py checks *.c files and
complains about violations of our intended #include orders.

Getting the quoting right to pass along the empty string in SunOS
is not working.

The code path is tested in vsl_glob_test.

fnmatch(3) appears to know FNM_IGNORECASE but not FNM_CASEFOLD.

Don't include vrt.h after cache_varnishd.h, which includes cache.h.

The extra locking prevents the race where a request thread is setting
r2->cond = NULL and the session thread is testing r2->cond != NULL.

Fixes: #2434

Keeping these in struct sess would necessitate extra locking for h/2
where we have multiple reqs in flight concurrently.

Fixes: #2268

This moves the body status decision to h2_end_headers, to make sure that
it is only done once per request.

We also move the Cookie header concatenation to h2_end_headers, to avoid
stepping on any changes that may have taken place in VCL.

Fixes: #2305

This is part of the $ABI reorg.

Two static checkers and myself had problems understanding the old code.

Avoid an undefined behavior of memcpy() in _tr_stored_block().

Allegedly the behavior of memcpy() is undefined if the source
pointer is NULL, even if the number of bytes to copy is zero.

Trying to tackle #2460

for i < 0, rlen could underflow. We are safe because of the check for
i < 0 further down, so this change is just a minor cleanup.

Fixes #2444