deeqSeq proposal

[Andy Gill]

On Apr 7, 2006, at 3:59 AM, Rene de Visser wrote:

> Hello,
>
> As deepSeq has a non local effect, I think it requires a non-local  
> source transformation to implement it. One option would be for the  
> compiler to create a second deepSeq version of every function  
> definition.
>
> e.g.
>
> If the user defines a function f
>
> f x = g h x
>
> then the compile creates an additional function !!f
>
> !!f x = temp `seq` temp
>          where temp = !!g !!h x
>
> which uses the compiler generated functions !!g and !!h.
>
> It looks like library writers are increasingly doing this manually.  
> Creating a strict and non strict version of a number of the  
> functions provided. This would automate that.
>
> Rene.
>

It depend on the semantics of deepSeq. If deepSeq just performs seq  
on all constructors recursively, then
that can be implemented as a runtime primitive. If deepSeq is making  
all embedded partial applications
strict, then yes this might be a non-local effect.

What are the semantics of !!(\ x -> ...)?

I am calling for the version of deepSeq/strict that evaluates all  
thunks, but does not strictify the arguments
to partial application, because
  - I believe this is straightforward to implement
  - It does not change the semantics (any more than seq)
  - I will address the problems we are trying to solve by ad-hoc  
means at Galois.

Andy