On Tue, Oct 12, 2010 at 2:08 AM, Simon Marlow
<marlowsd@gmail.com> wrote:
Nothing has changed in this area that I'm aware of. Could you give more details?
Sure. I'm building darcs HEAD of the text library using both 6.12.3 and 7:
cd text
cabal install --enable-library-profiling
It gets built with -prof -auto-all (see ghc-prof-opts in text.cabal).
If you then go into the tests/benchmarks directory, you can build and run the Equality benchmark:
ghc -prof -auto-all -O --make Equality
(add -rtsopts if using 7)
Then run Equality on it:
time ./Equality text twl06.txt +RTS -P
I've attached samples of the .prof files generated by each version of GHC so you can see the rather dramatic difference. To my eye, it looks like something is quite wrong with the version from 7.