[Haskell-cafe] how to check thunk

Austin Seipp mad.one at gmail.com
Mon Jul 2 12:04:41 CEST 2012


Hi,

Just a word of note: a while back, I decided to take up maintainership
of Vacuum and some associated stuff. In the process of doing this, I
realized that the ClosureType code in vacuum may not accurately model
reality depending on the GHC version. In particular, the definition of
ClosureType in vacuum as currently released on Hackage is woefully out
of date with respect to Modern GHC. Matt Morrow it seems originally
just copied the definition wholesale when he developed it, but as GHC
has evolved, so too has the ClosureType definition. Vacuum has not
kept up.

I have a github repository[1] containing several cleanups of Vacuum
which I'm calling Vacuum 2.0, one of these cleanups being that the
ClosureType definition is automatically generated from the GHC
definition, and compiled-in accordingly. I have also dropped support
for older GHCs (although it could be re-added.)

In this case, based on the changes I made I believe it is fairly
unlikely that isThunk would turn up being wrong here. The THUNK*
definitions seem to have been relatively untouched over GHC's
evolution. But for more advanced functionality, it could possibly
result in Vacuum as it stands constructing an invalid heap
representation based on object misinterpretations.

I have been busy with a new job but I am not totally absent, and I
think I will take this opportunity to release v2.0 in short order,
which is mostly intended as a polish and maintenance release for
modern GHCs. In the future I would like to see vacuum deprecated and
move all its functionality into GHC - much of the code is deeply
intertwined with GHC's runtime representations and in some cases I
have seen code in the package that was literally copied verbatim from
the GHC sources itself, obviously dating several years. It's hard to
assess the validity of much of this code. All of this feels brittle
and makes me uneasy, and given the complexity of the package as it
stands, will make it extremely difficult to maintain as GHC moves
forward.

Sorry for the ramble, but I figured it might be relevant if people are
going to rely on this functionality.

[1] https://github.com/thoughtpolice/vacuum

On Mon, Jul 2, 2012 at 2:54 AM, Erik Hesselink <hesselink at gmail.com> wrote:
> There is also the 'isevaluated' package (which depends on vacuum, but
> seems to do something more involved than your code).
>
> Erik
>
> On Mon, Jul 2, 2012 at 7:40 AM, Chaddaï Fouché <chaddai.fouche at gmail.com> wrote:
>> On Mon, Jul 2, 2012 at 5:29 AM, Kazu Yamamoto <kazu at iij.ad.jp> wrote:
>>> Hello,
>>>
>>> Are there any ways to see if a value is a thunk or memorized?
>>> I would like to have a function like:
>>>   isThunk :: a -> IO Bool
>>>
>>
>> vacuum allow that and much more though I don't know if it still works
>> correctly on GHC 7.4. Anyway your isThunk is
>>
>>> isThunk a = fmap GHC.Vacuum.ClosureType.isThunk GHC.Vacuum.closureType
>>
>> (Feel free to arrange your imports to make that a bit more readable ;)
>>
>> http://hackage.haskell.org/package/vacuum
>> --
>> Jedaï
>>
>> _______________________________________________
>> 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



-- 
Regards,
Austin



More information about the Haskell-Cafe mailing list