These tests were relying on the old behaviour of -fno-code, and work
just fine without it.

I'm not entirely sure if the cygwin code is actually right (i.e. I'm
not sure what calling convention it uses), but it seems to work.

It seems to be scheduling sensitive, and sometimes diverges when given
only one core.

Fixes openTempFile001 for some system codepages, most notably 437 (US).

I've only set this to run with the "normal" way atm because it 
takes about 1.5 min to compile on my machine. SpecConstr blows out
the size of the core program to about 400k, which is probably
a good enough reason to have it in the testsuite.

It broke when the installation path contained a space