Local extra-dep should not be considered $locals

[ndmitchell]

Using stack installed with --git on Windows 10 64bit, version:

Version 1.6.0, Git revision 46121be1b96465f1164e3f84cafa19c7369da9cc x86_64 hpack-0.18.1

Given stack-bug.zip, that defines in the stack.yaml:

packages:
- location: foo
  extra-deps: true

ghc-options:
  $locals: -bob

When compiling foo I don't expect the flag -bob to be applied as foo is an extra-dep, and thus not a local package. However, when I stack build I get an error about:

ghc.EXE: unrecognised flag: -bob

[mgsloan]

extra-deps is perhaps not the best name, a bit of a historical artifact. It would be nice to improve the stack documentation to describe the terminology properly. Or, perhaps, change the behavior to be less surprising.

"extra-dep" essentially just means "not a target by default". Currently, they are always considered local packages. Once implicit snapshots is implemented, this wouldn't be the case and the local package vs snapshot package will make even less sense. So I anticipate that the local package distinction will get dropped.

I think $targets is closer to the behavior of what you're looking for, but it will also be applied in the case that you do stack build foo, but won't be if you just do stack build.

Not sure what to call the non-extra-deps. default-targets? I guess could add another set of packages specified by the config, based on not being an extra-dep.

[ndmitchell]

Hmm, extra-deps is indeed a terrible name - we've used it completely incorrectly. Can I suggest adding definitions of $locals and the other $variables to yaml_configuration.md. Note that defining custom ghc-options makes something a $locals (I think). I did read the docs, and came to the conclusion (perhaps from extra-deps elsewhere on the page) that the behaviour is wrong.

FWIW $default-targets was exactly what I wanted, but following #3573 it's not as important, and given the extra-deps terminology elsewhere $default-targets would be reasonably confusing.

[snoyberg]

First things first: the correct syntax is to say extra-dep: true instead of extra-deps: true. In fact, if I run stack build in this directory, I see:

Warning: /Users/michael/Desktop/stack.yaml: Unrecognized field in PackageEntry: extra-deps

Changing to extra-dep: true gets me:

Error parsing targets: The project contains no local packages (packages not marked with 'extra-dep')

And then stack build foo gets the error in question.

I think this is just a bug in implementation: I would not consider any extra-dep packages to be locals, and think it's just a bug in the code base. My understanding and reading of the docs have always gone in that direction. @mgsloan are you inherently opposed to just fixing the implementation such that $locals only applies to actual local packages?

[mgsloan]

@snoyberg I am not opposed to that. I think part of the root of this is that IIRC, the LocalPackage type is also used for git packages, I suppose since the code is always available locally. It might actually be nice to reduce the distinctions between snapshot and local packages, keeping track of unpacked snapshot source. I don't think this would work for wired in packages / other packages that come with ghc, so that could be tricky.. It'd have the following advantages:

• Potentially implement Allow GHCi to load non-local packages by unpacking them #1441 without needing to do an unpack
• Would make it a bit easier to support "jump to definition" for dependencies in intero https://github.com/commercialhaskell/intero/issues/231

In fact, I think it would be good to take this as an opportunity to consolidate / define stack's terminology around different sorts of packages. Currently we have: the following terms

• snapshot package: this is usually pretty clear, but can get blurry when it comes to extra-deps
• extra-dep package: I interpret this as meaning "non default target". It's a bit of a weird name imho, certainly used to it by now.
• local package: ???
• target package: specified on the CLI, or implicitly selected.
• remote package: I think this means packages from repos? You'd think local vs remote would be antonyms, but they don't seem to be. I suppose that's the crux of this issue.
• hackage package: really means "from the package index / repository"
• stackage package: means a package from a stackage snapshot. Something currently tracked by stackage
• wired-in package: built into ghc, present in global DB

Would also be good to include along with the docs for this terminology, also the terminology for DBs, even though that's an implementation detail. It would be good to explain global DB vs snapshot DB vs local DB.

[snoyberg]

I think we're mostly on the same page, see #3352. For now, I'm going to make changes necessary so this specific case passes.

[snoyberg]

PR #3601 is now open

[snoyberg]

This was already addressed by #3601, closing.