My current speed-testing validating technique
igloo at earth.li
Sun Aug 12 17:58:19 EDT 2007
On Sun, Aug 12, 2007 at 02:43:20PM -0700, Stefan O'Rear wrote:
> On Sun, Aug 12, 2007 at 09:00:34PM +0100, Ian Lynagh wrote:
> > On Sun, Aug 12, 2007 at 08:37:44PM +0100, Simon Marlow wrote:
> > >
> > > Aargh! That's the data from ghc-inplace, not ghc itself. Since
> > > ghc-inplace became a Haskell program, it's harder to get it to recognise
> > > +RTS options. Supposedly ghc-inplace passes the +RTS options through to
> > > ghc itself, but for some reason -sstderr isn't getting through (perhaps
> > > others aren't getting through, we should investigate).
> > It is, but ghc-inplace also sees it, so we get both reports:
> > $ compiler/stage2/ghc-inplace +RTS -sstderr 2>&1 | grep allocated
> > 64,907,200 bytes allocated in the heap
> > 46,968 bytes allocated in the heap
> > Hmm, perhaps we need a better hack.
> How about bypassing shutdownHaskellAndExit, perhaps by some variation on
> foreign import unsafe ccall "stdlib.h _Exit" exitNow :: CInt -> IO ()
> (System.Posix.Process.exitImmediately would also work, but I suspect GHC
> runs on more non-POSIX systems than non-C99 systems.)
I meant for all flags, not only this one. Maybe we should just use a
small C program instead.
More information about the Cvs-ghc