I see. Thanks for the links, Greg Morrisett's notes are a great survey of
possible approaches to this problem.
It is funny because I had also come to the conclusion that JIT was the
neater solution, as long as JIT is available with the runtime... and you
don't have to implement it -- I guess it's
On 7/25/14 8:26 PM, Patrick Walton wrote:
Uniform value representations work well too (as
OCaml shows), but of course you'll pay a performance cost for that.
Oh, note that Greg's notes are a little bit out of date when discussing
the performance tradeoffs of uniform value representation. On
Hi,
Could you provide a link to Patrick's description of size/alignment-passing
implementation? I'm interested in these things.
Well, there could be a warning if the compiler switches to such an
implementation. It's arguably still better than not compiling at all.
However, I don't have enough
On 24/07/14 11:59 AM, Lionel Parreaux wrote:
I can't pronounce myself about the suitability of features in the Rust
language, but it may be worth noting that some convenient high-level
features are already present in the language, like garbage collection.
There isn't an implementation of
Hi,
So traits seem to be quite similar to Haskell's classes, being also used
for parametric polymorphism. Now, Haskell classes are usually implemented
using runtime dictionary passing. In general, code cannot be specialized
for every function call, since there may be an unbounded number of
You can avoid monomorphization by using trait objects, which erase
the precise implementing type through a vtable + pointer.
http://doc.rust-lang.org/tutorial.html#trait-objects-and-dynamic-method-dispatch
has some documentation.
On Tue, Jul 22, 2014 at 10:16 AM, Lionel Parreaux
this remindes me of the issue i got when trying to implement finger
trees in Rust so long ago
https://github.com/rust-lang/rust/issues/8613
I suggested to let add a way to specify (in the code) how match
functions do we want to generate and failing at runtime when the limit
is reached. This made
On 7/22/14 10:16 AM, Lionel Parreaux wrote:
I'm not sure whether this is a big problem in practice, but I was
wondering if it would be possible to switch to some runtime mechanism in
cases like this. Maybe we could make a special version of every generic
functions, that takes a dictionary at
