CPP macros suitable for partial evaluation

[andreasabel]

I looked into tools that could simplify my codebase by partially evaluating CPP conditionals.
E.g. if I have

#if __GLASGOW_HASKELL__ < 806
...
#if !MIN_VERSION_base(4,11,0)
...

and I no longer support GHC < 8.6 or base < 4.11 then I can get rid of some cruft by partially evaluating the conditions and throw away the conditional together with the branch that is now impossible..

I checked some tools that do some partial evaluation of CPP conditionals:

• unifdef (shipped on my system macOS): https://dotat.at/prog/unifdef/
• cppp: https://www.muppetlabs.com/~breadbox/software/cppp.html
• coan: https://coan2.sourceforge.net/index.php

These tools typically only support setting a symbol to a value, like __GLASGOW_HASKELL__=806. They do not accept predicates, i.e., I cannot set MIN_VERSION_base(4,11,0) to be true.

So if Cabal defined 3 separate symbols to specify a version, instead of a ternary predicate, this would be compatible with such tools.

MAJOR1_VERSION_base=4
MAJOR2_VERSION_base=11
MINOR_VERSION_base=0

This would be usable in Haskell code as such:

#if MAJOR1_VERSION_base >= 4 && MAJOR2_VERSION_base >= 11
...

(One could think of other macro names, like the triple VERSION_EDITION_base, VERSION_MAJOR_base, VERSION_MINOR_base---although the terminology edition for the "major-major" version is not established.)

[phadej]

Few comments, it's also GHC which defines MIN_VERSION_base macros, e.g. given Ex.hs

{-# LANGUAGE CPP #-}

main :: IO ()
#if MIN_VERSION_base(4,19,0)
main = print True
#else
main = print False
#endif

we have

% runghc-9.8.1 Ex.hs 
True
% runghc-9.6.2 Ex.hs
False

---

I don't know (or remember) why GHC defines them. Would be get to find out, and check if you'd also like/need to change GHC.

---

I'd prefer

#define SUPER_VERSION_base 4
#define MAJOR_VERSION_base 19
#define MINOR_VERSION_base 0
#define PATCH_VERSION_base 0

just because I likes justified identifiers :P

---

I'd appreciate if changes to cabal_macros.h generation were explicitly part of the .cabal format spec, i.e. to get new functionality, you'd need to use more recent cabal-version.

[phadej]

And not having new features of cabal_macros.h guarded behind cabal-version: ... will result into issues like #9331

This is a reason why I think that cabal_macros.h, Paths_pgkname.hs generating code SHOULD be in Cabal-syntax, as it's the package defining (most of) the file format.

Specifically, if something in Cabal-syntax changes (say, new field) then there probably should be a new cabal-version, and perfectly if we won't need to introduce new cabal-version: (CabalVersionSpec constructors) if nothing changes in Cabal-syntax. The latter is exactly what happened leading to #9331: No change in Cabal-syntax, so no-one thought about having new cabal-version nor cabal-version checks.

[michaelpj]

I was able to make this work by:

1. Replacing predicate calls by constant macros.
2. Using coan (I tried unifdef but it doesn't simplify conditional expressions)

In the end I did something like this to remove all MIN_VERSION_ghc(9,2,*) macros:

rg 'MIN_VERSION_ghc' -l | xargs -I {} bash -c "sed -i -e 's/MIN_VERSION_ghc(9,2,[0-9]*)/ALWAYS_TRUE/g' {} && coan source -r -DALWAYS_TRUE=1 {}"

[michaelpj]

I couldn't figure out a nice way to deal with __GLASGOW_HASKELL__ < 902, though. Setting it to a particular value would also replace all other __GLASGOW_HASKELL__ comparisons.

[michaelpj]

The same trick with replacing particular comparisons with a constant would work, though, if the form is consistent enough in your codebase.