PUBLIC

 

Never mind, turns out this was because OmitInterfacePragmas is turned on by default… (why?). After unsetting OmitInterfacePragmas in the dflags, I get the expected unfolding, and inter-module inlining works.

 

Unf{Src=InlineStable, TopLvl=True, Value=True, ConLike=True,

    WorkFree=True, Expandable=True,

    Guidance=ALWAYS_IF(arity=1,unsat_ok=False,boring_ok=False)

    Tmpl= \ (ds [Occ=Once1!] :: A) ->

            case ds of {

              A1 -> B1;

              A2 -> B2

            }}

 

 

 

PUBLIC

 

Hi,

 

The attached program uses the GHC 9.0.1 API to typecheck and compile a single module that contains a single definition marked as INLINE:

 

module Inline where

 

data A = A1 | A2

data B = B1 | B2

 

{-# INLINE f #-}

f :: A ->B

f A1 = B1

f A2 = B2

 

In my bigger program, I noticed that usages of `f` in another module aren’t inlined, so I made the attached experiment: I compile this to desugared, simplified, tidied and prep’d Core, and then use updateModDetailsIdInfos to fill in the unfolding of `f`. At least, that’s what I thought I do. However, if I print the unfolding of `f` after this process, I get `OtherCon []` instead of what I expected to be a CoreUnfolding. What am I doing wrong?

 

Thanks,

            Gergo