
Gabor PALI writes:
On Tue, May 18, 2010 at 11:10 PM, Ashish SHUKLA
wrote: I hope this was the only hurdle we'd. Now any ideas how to sanitize the environment ? Using env(1) or any other better way ?
Yeah, I remember this nice error message :) I think the problem was around an uninitialized variable (describing packages) in the Makefile. I have managed to fix it back in January, but I do not know where I left the solution :P Anyway, if you find something to patch, it would be nice to post the cvs-ghc.
LoL, and WTF, you should have at least posted it on the list. Anyways, I ain't sure if patching the ghc.mk to initialize PACKAGES is the right way to fix error. But anyways, for now I'm going to patch our Makefile to initialize that variable on the command-line.
I do not know exactly how to setup a Tinderbox-like jail (perhaps you should consult the sources), but GHC 6.12.x and 6.13 builds without any problems in a clean jailed environment for me. It seems Tinderbox is very picky... which is good for us in the long run. I solution would be to use the "enterBuild" feature of Tinderbox (I think I did that in January to hunt down the problem) but it requires root privileges.
Yes, I just looked at the differences in the *environment* of tinderbox, and there I noticed that PACKAGES is being initialized with '/tmp/packages'. Anyways, lets hope everything builds fine now. -- Ashish SHUKLA Sent via Gnus from GNU Emacs