[Haskell-cafe] ANN: Leksah 0.10.0

Daniel Fischer daniel.is.fischer at googlemail.com
Thu Apr 28 02:00:49 CEST 2011


Good thing I didn't send too soon this time:)

On Tuesday 26 April 2011 02:00:32, jutaro wrote:
> 
> Please try to run Leksah with the default config
> (~/.leksah-0.10/packageSources)
> 

> 
> Indeed leksah may use more memory on the first run (actually it is ghc,
> which uses it).
> But on consecutive starts it may use about/up to 150MB, but not the
> numbers you give.
> So please try to run Leksah with the default config, and see if the
> problem remains.

I started with:

"Worked like a charm. Generating the metadata on first start max'ed at 
about 260MB, which seems reasonable. On the second start - without 
[de]installing any package - leksah came up almost immediately"

And that's true so far, but the thing is, I hadn't yet installed all 
packages I had before.

So I installed the remaining packages (criterion and some of its 
dependencies) and tried again. First, to see whether my borked OS had 
anything to do with it, I moved ~/.leksah-0.10, started leksah and again 
pointed it to ~/.cabal...; again used very much memory, swapping, 
thrashing, kill -KILL. So it was not the OS (as expected). Then I deleted 
the new ~/.leksah-0.10, started leksah, this time with the default config. 
The same, swapping, thrashing, kill -KILL.

Hm, perhaps too many packages at once? What if we give it fewer?
So, delete ~/.leksah-0.10, copy the old one, start again. No dice.

Time for sleuthing. Unregister the new packages, delete .leksah and copy 
old, start leksah. Quick and easy.
Install stuff package by package. erf, fine. data-accessor, fine, ... . 
Chart, fine. statistics - boom. It grew to 1435m virtual memory, resident 
changing between 430m and 800m, constantly swapping, showed no sign of 
finishing after about five hours at 0-3% CPU with occasional bursts of a 
couple of seconds at >50%.

A chattier version of leksah-server indicates that it hangs on 
Statistics.Function.partialSort, but if debugM writes are buffered, it 
could well be something else.

So, is it just me or is statistics generally toxic for leksah?

> 
> Jürgen

Another problem: leksah.org isn't reachable anymore, 87.230.23.84 now 
resolves to a homoeopathist.



More information about the Haskell-Cafe mailing list