+RTS .. -RTS sections in ghc-options shouldn't affect the artifact hash

[23Skidoo]

If I build a project with new-build and then do new-clean && new-build --ghc-options="+RTS -A32M", it will try to rebuild all dependencies as well. This is not necessary and shouldn't happen.

[23Skidoo]

Related: #4247.

/cc @merijn

[merijn]

Actually, I don't think this is a matter of "RTS options shouldn't affect the hash" as, for executables, they definitely do affect the build output. This actually seems more related to the matter is discussed with @ezyang at some point on whether commandline flags (like --ghc-options) should by default be scoped to everything or just the current package (my argument then and in some ticket I can't find) was that commandline flags should default to only the current project and not all dependencies.

I'm pretty sure there's already a feature request for this somewhere and that'd obsolete the entire difficult task of somehow distinguishing some RTS options as safe and others as not.

[23Skidoo]

Actually, I don't think this is a matter of "RTS options shouldn't affect the hash" as, for executables, they definitely do affect the build output.

Hmm, I don't see how they do. This is about RTS options passed to the GHC executable itself via +RTS (e.g. to make the GHC process use a bigger GC nursery during compilation), not -rtsopts (default RTS options for the resulting exes).

Though if -rtsopts affects only exes and not libraries (I think it affects foreign libraries as well), it'd be nice if it didn't cause libraries to be rebuilt.

[merijn]

oh, I misinterpreted the problem! This should actually be fairly trivial. It's simply a matter of modifying normaliseArgs (

cabal/Cabal/Distribution/Simple/Program/GHC.hs

Line 49 in 86dabda

normaliseGhcArgs :: Maybe Version -> PackageDescription -> [String] -> [String]

) to strip out everything from +RTS to -RTS. But I don't have time to implement that right now.

If anyone else takes it up they should dive into the GHC 8.6 code and check if any flags were added/changed while they're at it, so the supported GHC range can be expanded to include 8.6.1.

[merijn]

@23Skidoo Should work properly with the PR I just created, but I don't have time to do a lot of testing, so please try dogfooding that one.

[23Skidoo]

Great, thanks, will do.

[23Skidoo]

I tested #5589 with GHC 8.4.1 on a private project, and it seems to work.