closing backend connections enough to make previously almost
stable test-cases more unstable.

Fix this one by having the server send Connection: close as appropriate.

there is insufficient workspace

Spotted by:	fgs

Reported by wouter-dot-hummelink-at-kpn-dot-com.
Reword slightly while I'm here. This could use some more love.

"feature pcre_jit".

This works (for test-case 1576 at least) on a 16k stack
and by default we have a 48k stack.

Augment the param description with some helpfull pointers.

the necessary #ifdef for the sparc64/32bit_userland thing.

Gcc is really, really weird...

We will not be able to put the more complex and compound parameters
into this table, but it will help keep types in struct param and
tweak_functions in sync.

Feel free to help get through the list.

(even if we exec env python, we should check early if we got one)

Instead of trying to steal vbc's away from the waiter, send the
request and wait for the waiter to hand the vbc back to us.

All waiters but poll still disabled ... coming up next.

sparse fd-sets efficiently.

to ensure this order

Don't worry, they'll will come back later.

and the ones which deal with fd's being waited on (Wait_*)