use of alex-3.5.2.0 on GHC/compiler/Lexer.x leads to errors

[shayne-fletcher]

the generated Lexer.hs fails to compile with

compiler/GHC/Parser/Lexer.x:3471:12: error: [GHC-31744]
    Duplicate INLINE pragmas for ‘alexScanUser’
    at _build/stage0/compiler/build/GHC/Parser/Lexer.hs:1294:12-23
       compiler/GHC/Parser/Lexer.x:3471:12-23
     |
3471 | -- in this file, alexScanUser gets broken out into a separate function and
     |            ^^^^^^^^^^^^

to reproduce

git clone https://gitlab.haskell.org/ghc/ghc.git --recursive
cd ghc
./boot && ./configure
./hadrian/build --flavour=quickest -j

[andreasabel]

Thanks for the report.

Duplicate INLINE pragmas for ‘alexScanUser’

GHC should not fail hard if someone passes the same information twice instead of just once.
(Haven't investigated if this has been reported as issue there already.)

I am open for discussion on whether the INLINE pragma now generated by Alex is a good or bad idea.
Apparently the GHC folks want it, and since Alex didn't do it by default they added it in.

CC: @brandonchinn178

[brandonchinn178]

I figured it would be fairly uncontroversial; I can't imagine a situation where you wouldnt want to inline it. I didnt realize that GHC would fail if it's passed twice. I'm open to reverting it to avoid a breaking change, but IMO this is something that belongs in Alex, not user-code

[andreasabel]

@brandonchinn178 From looking at the GHC sources, it seems like you introduced the INLINE alexScanUser there.
(Their issue/PR 25158.)

-- If the generated alexScan/alexScanUser functions are called multiple times
-- in this file, alexScanUser gets broken out into a separate function and
-- increases memory usage. Make sure GHC inlines this function and optimizes it.
{-# INLINE alexScanUser #-}

Would it be possible for you to place this pragma there under a suitable#if to reconcile it with alex-3.5.2.0?

• https://gitlab.haskell.org/ghc/ghc/-/issues/25623