| However, my proposal is not anywhere fundamentally difficult -- in its essence, I just | propose to move the implementation of the thread allocation strategy from the RTS/C | code, to a Haskell library. This gives programmers both a low-level interface for | explicit access and a high-level interface as it is now. It may well not be difficult, but nevertheless I am having difficulty understanding it. I'll have to wait till you have time to explain it. | What I mostly wanted to ensure is that people have really thought about this carefully | and that they could give strong reasons for choosing a particular design over another. The difficulty is that I can't give strong reasons for choosing X over Y when I don't understand what Y is. Next week is ok though. Good luck with thesis writing! Simon
Daan writes:
What I mostly wanted to ensure is that people have really thought about this carefully and that they could give strong reasons for choosing a particular design over another.
SimonPJ replies:
The difficulty is that I can't give strong reasons for choosing X over Y when I don't understand what Y is. Next week is ok though.
I think the essence of Daan's proposal was that the goals we hope to achieve using 'bound', 'threadsafe', and other ffi annotations could be achieved by adding a small amount of additional functionality and wrappers and that benefits of doing this are: - greater transparency to programmers - greater flexibility because those primitives can be used in many different ways - more future-proof because of greater flexibility - simpler design Some inspiration for this approach might be found in the move from GreenCard to the standard FFI. GreenCard gives you some important functionality that covers many cases you might want but, since it doesn't cover _everything_ you want to do, we've seen a gradual addition of features over the years as the shoe is made to pinch a little less. I think the details of which operations to add were largely intended to show that this kind of approach was plausible rather than being intended as a complete proposal. [Personally, I'm not sure that Daan is right - but I think it's worth exploring a little.] -- Alastair Reid alastair@reid-consulting-uk.ltd.uk Reid Consulting (UK) Limited http://www.reid-consulting-uk.ltd.uk/alastair/
participants (2)
-
Alastair Reid -
Simon Peyton-Jones