Hi again, no real progress, but a few small points (--buildwith=gcc): 1) what exactly is "fullname" supposed to do in configure? if fullname 2>>/dev/null # cope with symbolic links in directory paths then echo "$BUILDBINDIR/hmake-config $BUILDLIBDIR/$MACHINE/hmakerc \ add `fullname ./script/nhc98`" \ >>targets/$MACHINE/hmake3.config fi; And where is it supposed to come from? I can't find it, and neither can configure: Updating targets/ix86-CYGWIN_NT-5.0/hmake3.config... fullname: not found Saving current configuration in targets/ix86-CYGWIN_NT-5.0/config.cache Done. 2) is this really the line you want to get (in MkConfig.hs:210)? $ grep 'NHC98INCDIR' /tmp/nhc98-1.12/script/nhc98 | cut -c27- | cut -d'}' -f1 | head -1 find the hi files for the standard prelude, The path appears to be the second line in the output, just after the comment line;-) 3) after telling hmake-config not to bother with ghc, the case of the failing write to /tmp/.. appears to be in the second call to runAndReadStdout in MkConfig.hs (see 2), and all those calls use the same tmpfile for output from "system" calls. If the "grep" is the second call, that means that one call has succeeded without permission problems. That looks a lot as if "noclobber" was set in the system shell (causing output redirections to existing files to fail), only that if I just ask my toplevel bash, it claims that noclobber is off. Does system itself do anything fancy here (probably not)? If I wouldn't be compiling from .hc-files, I could just play with MkConfig.hs (using a different tmpfile for each call, or removing the tmpfiles after use, or even using ">|" for redirection to override noclobber). And I can't just run MkConfig.hs through hugs or ghc, because they would both defer system calls to the windows-shell, not to a version of sh, as the cygwin-based nhc will do, right? Is there an easy way for me to generate MkConfig.hc from MkConfig.hs, so that I can do some debugging? Any other ideas? Anyone else in the game?-) Claus