ghcversion.h

[hsyl20]

The RTS provides ghcversion.h which contains:

#if !defined(__GHCVERSION_H__)
#define __GHCVERSION_H__

#define __GLASGOW_HASKELL__ 913
#define __GLASGOW_HASKELL_FULL_VERSION__ "9.13"

#define __GLASGOW_HASKELL_PATCHLEVEL1__ 0
#define __GLASGOW_HASKELL_PATCHLEVEL2__ 0

#define MIN_VERSION_GLASGOW_HASKELL(ma,mi,pl1,pl2) ( \
   ((ma)*100+(mi)) <  __GLASGOW_HASKELL__ || \
   ((ma)*100+(mi)) == __GLASGOW_HASKELL__    \
          && (pl1) <  __GLASGOW_HASKELL_PATCHLEVEL1__ || \
   ((ma)*100+(mi)) == __GLASGOW_HASKELL__    \
          && (pl1) == __GLASGOW_HASKELL_PATCHLEVEL1__ \
          && (pl2) <= __GLASGOW_HASKELL_PATCHLEVEL2__ )

#endif /* __GHCVERSION_H__ */         

(adapted for the GHC version of course)

The issue is: if we try to use GHC without having a rts in the package db, then it fails to find this file and crashes.

We can force GHC to use another ghcversion.h file by using -ghcversion-file=foobar/ghcversion.h.

However cabal uses GHC to build a dumb C file in configureLd to test if ld -x works. But it doesn't pass -ghcversion-file and if the package db is empty it crashes.

Solutions:

1. don't crash, even if no ghcversion.h file is found, just display a warning. In https://gitlab.haskell.org/ghc/ghc/-/commit/12a7444463184e9eddbe7b7251a0ee1e976f4d75 @angerman wrote that it is unlikely that someone would use GHC to build a C file like this, but here we are.
2. allow GHC to generate this header file on the fly! The compiler knows its own version. Why is it in the RTS to begin with?

[hsyl20]

I've pushed a patch implementing solution 1 in #3

[angerman]

@hsyl20 I completely agree with your assessment in 2. I think having 1+2 is good.
Why this file lives in the RTS makes no sense to me, it's a compiler file, it should probably live in the compiler package it corresponds to? And as such should either be installed into
lib/includes (e.g. ghc will just look at $ghc/.../lib/includes/ghcversion.h) or as you say, just generate it on the fly, that however might be a bit annoying with temporary directories and files being created?

[andreabedini]

I have implemented generating ghcversion.h at GHC runtime (i.e. when compiling a module). The patch is 0f619b7.

[hsyl20]

Nice! It could be fixed directly upstream imo. Perhaps worth opening a ticket

[hsyl20]

I've opened the upstream ticket https://gitlab.haskell.org/ghc/ghc/-/issues/26018 and MR https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14300 with @andreabedini's patch adapted and with a test.

[hsyl20]

Fixed upstream.