DEV Community

Improving Elm's compiler output

Robin Heggelund Hansen on May 25, 2019

Elm is fast. This is not because of any innovation in the compiler. In fact, the Elm compiler hardly does any optimization at all. Elm is fast be...

Read full post

Gage • May 26 '19

Awesome changes! Are these changes going to be implemented soon?

Robin Heggelund Hansen • May 26 '19

The honest answer to that is: don’t know 😅

Gage • May 28 '19

Corollary, are you helping Evan directly now?

Robin Heggelund Hansen • May 29 '19

I'm a member of the core team, so I do talk with Evan from time to time. But I still write PRs which he may, or may not, merge.

Gage • May 29 '19

I just thought it was a core team of one, so that's cool 😁

Robin Heggelund Hansen • May 29 '19

There's actually a whole bunch: elm-lang.org/community

James Carlson • Dec 27 '19

Wonderfully informative post.

m-shaka • Dec 23 '20

Interesting article! Can I translate it into Japanese?

Robin Heggelund Hansen • Dec 27 '20

Sure :)

m-shaka • Jan 3 '21

I'm sorry there is something that I can't understand in Inlining section.

Is the intermediary number value is accumlator?
What does we were folding over other things mean actually? And why aren't ints stored? In my knowledge, all local variables are stored on the stack or the heap.

I'm looking forward to your answer

Robin Heggelund Hansen • Jan 3 '21

yes
other things like Booleans, Maps, Sets, Floats or other Objects. «If we weren’t storing ints» refers to the option of storing other intermediate values like objects.

m-shaka • Jan 3 '21

Thanks for your quick response! I understood.

I'm so sorry I have one more question, but this is last one.

It's about this sentence.

This is like getting the benefits of a monomorphising compiler, without actually having a monomorphishing compiler.

What I understand is below:
This talks about Inline Caching. If foldl is inlined, the function as argument of foldl is also inlined, so without Inline Caching code of inlined foldl's loop is specilized for argument functions type.
Is this right?

Robin Heggelund Hansen • Jan 4 '21

This talks about Inline Caching => inline caches is something else, inlining is just replacing a function call with the implementation of the function.

without Inline Caching code of inlined foldl's loop is specilized for argument functions type => without inlining, the foldl's loop cannot be specialized because it has to handle all possible argument types.

m-shaka • Jan 4 '21

Thank you! You mean monomophishing is to specialize code by inlining function call, right? If so, I can't understand what without actually having a monomorphishing compiler means...

Robin Heggelund Hansen • Jan 5 '21

monomorphising doesn't do inling. When a monomorphising compiler sees the following code:

listmap : (a -> b) -> List a -> List b
listmap fn list =
    List.foldl (\a acc -> fn a :: acc) list
    |> List.reverse

strs = listmap String.fromInt [ 1, 2, 3 ]

strs = listmap isEven [ 1, 2, 3 ]

it will generate two versions of listmap, one which has the type (Int -> String) -> List Int -> List String and one which has the type (Int -> Bool) -> List Int -> List Bool. It will then use the specialized listmap implementations where it can.

This allows the compiler to specialize code better, because it has precise type information.

m-shaka • Jan 5 '21

Oh...I misread that you used "monomorphishing" like as this article explained.

I've published my translation! zenn.dev/mshaka/articles/633027bef...

I appreciate your patience and kindness!