On 02 August 2006 19:09, Duncan Coutts wrote:
On Tue, 2006-08-01 at 12:12 +0100, Simon Marlow wrote:
Duncan Coutts wrote:
Specifically: if the user asks for profiling but not vanilla versions of libs and they're using TH then we should build vanilla libs first and then profiling libs. However at install time we would not install the vanilla libs as the user did not ask for them.
Yes, that's right.
Simon: what about if we're building a program rather than a lib? Should we vanilla-compile all the modules but not link then into a binary and then compile again with profiling and then link those into a binary?
I would think so, yes. Just build with --make -c for the vanilla way.
Ok I've got a patch attached that does this.
And presumably we would only expect this to work with GHC 6.6.
Yes. I'd appreciate it if someone could test this with a TH program and 6.6, to make sure that GHC can find the vanilla object files correctly.
Should I expect this to work now or is it still unfinished work for 6.6 ?
Yes, it should work now. I've turned on the profiling way for our TH tests, and they are passing right now.
I've added a trac ticket so we don't forget the testing:
Which reminds me of a question I have in general: what's the testing strategy for Cabal? I'm aware of the unit tests (which I use via GHCi). 'make moduleTest' spews a ton of stuff, including lots of messages about failures, and I'm not sure whether I should expect it to work or not. In GHC we have a small number of tests in our testsuite for Cabal-related things (http://darcs.haskell.org/testsuite/tests/ghc-regress/cabal). Perhaps we should try to incorporate the other Cabal tests into there too? Cheers, Simon