Note this could also be done slightly differently, making the -optl=-static flag that we're passing to GHC here more generic.

As per man ld, the -static and -shared can be passed to the linker together, which

means that a shared library is being created but that all of the library's external references must be resolved by pulling in entries from static libraries".

That means it might be sensible in some use cases to pass -optl=-static when we are building a Haskell shared library (on Linux that is building an .so file that doesn't depend on any other .so files).

If we want to support that case, then maybe we don't want to call this feature --enable-executable-static, but something like --enable-fully-static-linking instead and have it work for building static exes and ".so file that doesn't depend on any other .so files".

But that those aren't necessarily exclusive; Cabal may want to offer a high-level flag called --enable-executable-static anyway and a separate one to allow the ".so file that doesn't depend on any other .so files" use case.

Just some ideas in my head, what do you think?

CI should agree though.

• I am unsure whether I should also pass -static in GHCJS.hs as I do in GHC.hs

Who can tell me?

I have added the corresponding cabal-install parts, please take another look.

Some gold tests seem to fail.

How can I run them locally?

The travis output on https://travis-ci.org/haskell/cabal/jobs/406458613 is close to useless.

The raw output at https://api.travis-ci.org/v3/job/406458613/log.txt has lots of garbled characters like �[0Ktravis_fold:end:before_cache.2 and seems to contain the actual test case failures like unrecognized 'configure' option --disable-executable-static'` somewhere in the middle.

-static + -shared sounds like it would be good for foreign-libraries, which as root package / final artifact should be controlled with --enable-executable-static despite the weird naming.

GHCJS doesn't have alternate linking, so it really shouldn't matter.

I think this is the 4th time it complains about ChangeLog.md being conflicting because somebody else has pushed a change meanwhile. I am getting tired of needlessly rebasing due to this.

I've filed #5625 for it.

@nh2 I would recommend ignoring this by default on macOS. While its theoretically possible to build completely static macOS binaries, Apple doesn't provide anything like the stable ABI that Linux has. You can of course build all of your libraries statically but no matter what you will need to link against /usr/lib/libSystem.B.dylib. I am not sure what would happen when this flag is on in macOS? It might actually just ignore it.

I would recommend ignoring this by default on macOS

@matthewbauer Would there be a benefit of Cabal ignoring the flag though, vs just having users not use it on OSX?

Accepting the flag and then going not doing anything goes against the principle of least surprise, I'd rather have it fail (and put into the docs that it isn't known to work on OSX), or fail loudly that it doesn't work on OSX (and why).

I think this is the 4th time it complains about ChangeLog.md

and another time:

This branch has conflicts that must be resolved
Conflicting files
Cabal/ChangeLog.md

@23Skidoo I am unsure right now whether #5625 (comment) means that this PR will actually merge cleanly if not done via GitHub, or not.

I would recommend ignoring this by default on macOS

@matthewbauer Would there be a benefit of Cabal ignoring the flag though, vs just having users not use it on OSX?

Accepting the flag and then going not doing anything goes against the principle of least surprise, I'd rather have it fail (and put into the docs that it isn't known to work on OSX), or fail loudly that it doesn't work on OSX (and why).

I'd advocate for uniform behaviour: let the option work on all OSes, but add an explanatory warning on Mac. In other words: discourage over completely preventing.

@nh2 I tried rebasing, and it didn't require manual intervention, but put the changelog note into the 2.4.0.0 section, so it had to be fixed manually.

Is there anything that an interested passerby like myself could do to help get this merged?

Statically linked Haskell executables are something I'm very much interested, so I'd like to help unblock anything if it's at all within my capability 😄

@23Skidoo: What's holding this up from being merged?

@nh2 wanted to add some tests for this. I think I'll merge this anyway rn, but a subsequent PR with tests would be appreciated!

Merged, thanks!

Regarding tests, I'm still kinda blocked on

• would appreciate feedback where those should go; can I even do that in Cabal's test suite if I need e.g. musl to test it?

How would you want something like this to happen?

Feature detection, and skipping these tests when musl isn't available?

Yeah, that'd be fine. Can we install musl on Travis?

(I haven't done anything with Travis besides copy-paste-ing configuration files, but...) googling "travis musl":

travis-ci/travis-ci#6128

it sounds like there's no official support, but one can still just install a virtual machine with alpine linux (or some other environment with musl present).

i just set up Alpine Linux under Virtualbox on Ubuntu LTS 16.04 (afaik, that's the OS which Travis CI runs stuff on by default).

@nh2 do we need install to the cabal toolchain on it? or just check that any statically linked executables built by cabal ----enable-executable-static on the same platform (64-bit Intel, Linux) run (and/or pass some tests like ldd hello | grep "not a dynamic executable")?

@sboosali Hmm, I think you will need a full musl toolchain all the way up to ghc for it to work.

I would be surprised if running a VM on Travis's Ubuntu works well (their environments are already quite slow, and if they don't have nested virtualisation, it might be unbearably slow).

Another alternative might be to link statically with glibc, and pick a very trivial hello-world program that is unlikely to trigger any code paths that are problematic when linking glibc statically.