Stephan Brandauer
Stephan Brandauer
Before we can merge this, we need to (at least) train/evaluate/release the models, and add a commit here that makes the branch depend on the new model.
I just requested a review from @henrymercer - this is a pretty big change set. This is the branch we're aiming to release, as discussed yesterday evening. We've been using...
Experiment came back, 34% slowdown. @tiferet, is this a blocker?
> Is this a 34% slowdown relative to the unboosted version or relative to the current version of ATM? Relative to the unboosted version :) > Also, @kaeluka do you...
- I think that a version of your absolute paths proposal makes sense (doesn't mean it's the only one that makes sense), *but* I don't like that your examples contain...
- The form `expr1[expr2]` could clash with type parameter syntax: If `x` is an array of closures that take `int`: `x[0](123)` - The store expression should probably return `void`
An alternative would be to use call syntax: `x(0)`. It makes an array look like a function that accepts an index. The reason for `[]` is of course that it's...
> Re your second point, that's what I first thought too. But it seems strange that loops etc. have a return value and stores don't. I don't mean to sound...
The 'canonical' justification for making assignments void is that: `if (x = true) then ... else ... end` should not typecheck.
This typo is much more common in languages with weird semantics of booleans (like C), but the fact that it would be extra unlikely to happen in encore would make...