[Haskell-cafe] Re: A guess on stack-overflows - thunksbuild-upand tail recursion

Claus Reinke claus.reinke at talk21.com
Fri Mar 20 19:36:45 EDT 2009

> It would be great to have a video of this in action up on youtube.
> You can simply 'recordmydesktop' on linux (and likely elsewhere), then
> upload the result.

I'm curious: how would a non-interactive animation running in Flash
in a browser be better than an interactive animation running in Java
in a browser?-) When I wrote GHood (many years ago), I explicitly
looked into the applet option, in the hope that people would use it
to document and discuss observation logs of their favourite Haskell
strictness issues, with animations available on their web pages, right
next to the discussions. 

That hasn't happened yet (the only users I was aware of were the
DrHylo/Pointless Haskell project), but I just checked, the old .jar file,
the source of which hasn't been perused for a long time, still worked 
in applet mode (in Opera, a browser I didn't know about in 2001,
using a Java Runtime several versions removed from that time - try
that in Haskell.. ;-), straight from that old project page (which also 
explains how to set such things up), so anyone could add animations 
of their favourite examples on their web-pages. But don't let that 
keep you or anyone else from addressing the youtube audience 
(one could add audio explanations, I guess).


PS. Perhaps these days, someone should rewrite the log viewer
    in Canvas+JavaScript as a more lightweight and modern platform.

> It also helps the general adoption cause, having Haskell more visible
> and accessible.
> claus.reinke:
>>>> The problem occurs when the result value is needed and thus the   
>>>> thunks need to be reduced, starting with the outermost, which can't   
>>>> be reduced without reducing the next one .... etc and it's these   
>>>> reduction steps that are pushed on the stack until its size cause a   
>>>> stack-overflow.
>>> Yes, that's exactly right, and something that's not often pointed out.
>> Btw, this is kind of relative strictness (when is one part of my program
>> needed to answer demands on another part) is the kind of example
>> for which old GHood can be helpful (once you get used to the display).
>> If you have Java on your machines, try installing GHood [1] (on hackage 
>> thanks to Hugo Pacheco), then things like
>> ghc -e ':m +Debug.Observe' -e 'printO $ observe "foldr" foldr (+) 0 [1..4] '
>> ghc -e ':m +Debug.Observe' -e "printO $ observe \"foldl'\" foldl' (+) 0 [1..4] "
>> ghc -e ':m +Debug.Observe' -e 'printO $ observe "foldl" foldl (+) 0 [1..4] '
>> This was also among the examples on the GHood home page [2], so you could 
>> try the applet version instead, and in section 4.2 of the paper [3] (as a 
>> "well known strictness problem";-). Page and paper
>> mention a few other similar examples and discuss some differences
>> between static (which parts are needed at all) and dynamic strictness
>> (which parts are needed when, relative to other demands).
>> Claus
>> [1] http://hackage.haskell.org/cgi-bin/hackage-scripts/package/GHood
>> [2] http://www.cs.kent.ac.uk/~cr3/toolbox/haskell/GHood
>> [3] http://www.cs.kent.ac.uk/~cr3/publications/GHood.html
>> _______________________________________________
>> Haskell-Cafe mailing list
>> Haskell-Cafe at haskell.org
>> http://www.haskell.org/mailman/listinfo/haskell-cafe
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe

More information about the Haskell-Cafe mailing list