12:54 <aiowej> Hi! I want to learn how to implement a compiler for a language like ML, but purely functional, really nothing more than a typed lambdacalc with pattern matching and union types. I already know how to type check, but am stuck on how to create the actual output. Looking to output something like C++, Rust, or LLVM IR, i.e. no closures. Maybe should do CPS? Anyways, my question is this: what book do you recommend for someone who

12:54 <aiowej> just-wants-to-get-it-done, compiling a pure lambda language to C/LLVM?

13:01 <def`> Andrew Appel modern compiler implementation in ML? You can just skip the lower level parts.

13:02 <def`> (And imho, producing C++, Rust or LLVM IR will just make things more complex, stick to C, it is portable)

13:05 <aiowej> def`: thank you! I know his book Compiling with Continuations, which is cool, but being 1/3 through it I'm not sure it has all the answers (or I'm a dum dum). Will check the book out. There are some data structures implemented for both C++ and Rust I want in my stdlib

13:10 <aiowej> def`: if you have read both books, I would love to heard what valuable knowledge is available in Modern Compiler Implementation in ML as compared to Compiling with Continuation. Many thanks

13:11 <aiowej> :s/heard/hear

13:52 <Leonidas> wow, 'Illegal permutation of structure fields' when I attempt to PPX generate signatures :(

13:56 <Drup> eh, that's a fun one

13:56 <Drup> it's quite challenging to trigger

14:00 <def`> :D

14:12 <Leonidas> I managed to do that with a `module type` in an ml and the same `module type` in an mli

14:14 <Drup> exactly the same one ?

14:14 <Leonidas> no, the ml had the same sig defined a second time, redundantly

14:16 <Leonidas> I found an 18 year old discussion on caml-list about this error, quite impressive. Markus Mottl was around back then already

14:20 <octachron> Normally, module types cannot be defined twice?

14:22 <Leonidas> octachron: I am a bit annoyed by this myself, because it is now 100% code duplication. Maybe I should just move it into an _impl module

14:24 <octachron> You are generating the mli and ml files in completely different phases?

14:24 <Drup> that shouldn't be a problem, as long as they are the same

14:40 <Leonidas> octachron: no, my issue was that I had the type in the ml file with a manual definition, why in the mli I just let my deriving plugin generate it.

14:41 <Leonidas> and this seems to generate subtly different things since ppxlib.deriver wraps these generated signatures in a sig that it then includes ¯\_(ツ)_/¯

15:45 <spew> what ppx is this?

15:48 <Leonidas> spew: my own

15:49 <Leonidas> that I'm writing

15:50 <Leonidas> which is why I didn't use the signature generation, because I didn't trust it yet and thought I could go by with manually writing it, thus triggering the error

16:02 <spew> is it like https://github.com/ocaml-ppx/ppx_import ?

16:02 <spew> the thing you're writing?

17:46 <Leonidas> spew: no, it is, uhm, a bit like ppx_deriving_yojson just that it derives a schema from the type

17:47 <Leonidas> the schema in question is a bigquery table definition

18:05 <spew> neat