Compiling Functional Languages (2002) [pdf] | Hacker News Reader

Compiling Functional Languages (2002) [pdf]

xavierleroy.org

78 pointsperihelions14 comments

This presentation is very good! A few days ago, I started thinking how code from a functional language could be converted to imperative code for better performance, with in-place operations (instead of object regeneration) and a complex (and likely slow) optimizing compiler. However, things like closures, composition at runtime and partial function application just made everything 10x harder.

> However, things like closures, composition at runtime and partial function application just made everything 10x harder.

Closures are just functions with an extra parameter for the environment (free variables). Subsequent passes after desugaring can treat them like any other function.

Function composition and currying should also be static transformations. Maybe you made it harder than it has to be?

If functions are higher order then the distinction between plain functions and closures is retained (unless you treat all functions as closures). This is because the data needs to be passed around in the function pointer. If not higher order then yes, they desugar

Implementing these things is maybe easy for you ;) but for myself, doing it well, I would definitely not describe as a walk in the park

It is true you need to pass the environment around, but if the language you are implementing also has objects or structs then you will need to implement that anyway.

My main objection to the grand parent is the idea that functional languages are 10x more difficult to implement than mainstream languages. Maybe that is true for Haskell, but most functional features can be compiled efficiently without much additional complexity. The complexity added by closures for instance would be closer to 1% than 10x.

for real "fun", have a look at continuations :-)

That's the same curiosity that led me here: how much of higher-order functional programming could be implemented as zero-overhead abstractions over a C-like language, and what restrictions on language semantics would be the tradeoff?

State monad as a first class language construct might help? You could allow actual mutations inside that and therefore if a lot of computation needs doing it could be more efficient.

All of it. No restrictions on semantics. Existence proof - any compiler that has a first order IR. SML's mlton would be a good example.

"First order IR" in the sense of "without lambda abstractions"? Then I guess this [0] would be a decent starting point for understanding it? It seems to describe mlton's compiler pass for flattening the higher-order functional parts with some kind of flow analysis. (?)

[0] http://www.mlton.org/guide/20130715/References.attachments/C... ("Flow-Directed Closure Conversion for Typed Languages")

First order probably means no passing functions as values. That's somewhat imprecise, passing a function pointer to a function called call or apply seems fine. No passing closures around though.

One way to go is compilation to combinators, which are functions that don't have a runtime representation of a lexical environment, aka functions that can be serialised as C without change in semantics. Basically add arguments to represent closed over data and rewrite call sites.

On reflection I think continuation capture is a candidate for inherent overhead that cannot be eliminated - at least I can't currently see how to eliminate it - but then C can't do that. There's also the question of runtime cost of error detection, out of bounds etc, which C also doesn't do.

You probably need whole program compilation to desugar everything. And I'm essentially claiming a sufficiently smart compiler can make functional languages as fast as imperative, which has dubious empirical support. Stuff like hash tables are hard to express without mutation or overhead. Still, mlton and stalin (r4rs scheme) take a reasonable stab at it.

Probably fair to say that with today's compiler tech I'm wrong, but not in a fundamentally unsolvable tomorrow sort of way.

Great question

I wonder what is a zero overhead abstraction. For instance closures are expensive (they malloc), but to what extent can we consider closures to be malloc'd structs; then potentially do away with the concept of structs -- because they've become a subset of the functionality provided by closures. At that point can we not consider closures as zero cost; because malloc of structs is essential to c, and with optimization we need be no more expensive

And to what extent does that hold true for every other feature? In some sense it is maybe always true

Interestingly, the last slide mentions:

``` Today: we are able to certify realistic bytecode compilers and abstract machines.

Tomorrow: certification of optimizing native-code compilers? ```

Xavier Leroy actually got the ACM Software System Award in 2021 for CompCert, an optimizing native-code compiler! It's first version seems to have been released in 2005, just 3 years after this presentation was written. Though CompCert is a C compiler, which isn't really a functional language.

- https://en.wikipedia.org/wiki/ACM_Software_System_Award - https://en.wikipedia.org/wiki/CompCert

CakeML is a formally verified implementation of a subset of Standard ML which generates native code. Not sure about what it optimises though.

Xavier Leroy is one of the main developers of OCaml.

Reply on news.ycombinator.com