Re: loop performance bug

Hi again,

> > A first hint is to never try to guess what kind of code ghc generates.
> > If you're in need of performance you need to look at some lower level
> > code. I recommend the -fext-core flag to produce external core, a
> > sort-of-readable output of ghc's internal representation of the
> > program. One can learn a lot from that.
> 
> Although when I do I don't necessarily understand what I see! I see
> 'let' (ie allocating forms) where I would have hoped there would not be
> but it doesn't help me to figure out why the code I've written ends up
> producing allocations.
> 
Reading core programs can indeed be confusing and difficult. It's an
art that has to be mastered. All I can say is that in situations like
this it can be rewarding.

> > I'm very much surprised that this program was slower than the one
> > without arrays.
> 
> You mean the one without lists.
> 
Yes :-)

> >  Since all lists are produced and consumed with "good"
> > producers and consumers (as in section 7.10.3 in the User's Guide)
> > they should really go away. This seems to be a real bug.
> 
> This is what Simon M thought when I brought this up previously. He
> suggested I use explicit recursion (which I thought I was doing with my
> second version though it doesn't do much better!)
> 
And I suggest Simon M tries to fix ghc so that you don't have to write
the program using explicit recursion :-)

> I was hoping that if the array reading was inlined that the calculation
> of the index would also be inlined and so the Pos constructor need never
> be allocated.
> 
I think this is a fair assumption. But apparently ghc doesn't remove
the overhead. Maybe ghc's optimizer needs to be beefed up.

> BTW have you thought about picking up hIDE development again? I might go
> back to it after Gtk2Hs hits version 1.0
> 
You're confusing me with Jonas Svensson. I don't blame you, it's an
easy mistake to do if you're not a swede. So the answer is no, I'm not
going to develop hIDE any further :-) I should mention though that
there is a group of students here at Chalmers working on a Haskell
IDE. I don't know the status of their project though.

/Josef