[darcs-devel] Re: Building darcs

[David Roundy]

On Thu, Feb 08, 2007 at 11:13:49PM -0500, Zachary P. Landau wrote:
> Thanks for the tips.  When I disabled optimization and used -j2 I was
> able to get the build from about 9 minutes to about 3 minutes.  (I'm not
> sure why I got 15 when I was testing earlier, maybe I had something in
> the background that was using up resources).  3 minutes is short enough
> that I can stay focused and long enough to force me to consider if my
> code is correct, so that is probably good enough for me.

I meant to reply to this question earlier, but have been rather busy.  I'm
afraid I just wait while it compiles for a very long time (and try to make
all the changes to code that everything else depends on quite early in the
development cycle, if possible).  My home devel machine (where I do darcs
development) is a 700MHz PIII, so I can share your pain.

I do find that *usually* a longer compile time seems to help by giving me
more time to think.  But sometimes it certainly just gives me more time to
be distracted (particularly when the entire code base needs to be
recompiled).

And turning off optimizations can sometimes modify buggy behavior, if they
have to do with laziness, so you do want to be cautious with that while
debugging.

> That being said, I still wouldn't mind if I could load darcs through
> ghci when testing.  Is that just a pipe dream, given the dependencies
> that darcs pulls in and whatnot, or have others gotten it to work with
> recent versions?

I know someone got this working once.  Certainly we could easily load darcs
in ghci if we relied more heavily on packaged haskell packages rather than
doing so much ffi ourselves.  Also, turning off libcurl support, etc, would
help make that easier.  e.g. if/when we switch to using Data.ByteString, we
can then use the zlib haskell package rather than linking to zlib
ourselves.  This'll mean better tuned code and generally better behavior,
but will mean that our users (when compiling for themselves) will need to
download and install the zlib package as well as darcs itself.  Similarly,
it'd be nice to allow the use of a pure haskell http module, but that'd
also take some care in handling.  Once we've made some of those changes, I
don't see why using ghci wouldn't be trivial.
-- 
David Roundy
Department of Physics
Oregon State University
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.osuosl.org/pipermail/darcs-devel/attachments/20070209/f916c150/attachment.pgp