Make dot and ls dependencies work without ghc installed

[waddlaw]

Note: Documentation fixes for https://docs.haskellstack.org/en/stable/ should target the "stable" branch, not master.

Please include the following checklist in your PR:

• Any changes that could be relevant to users have been recorded in the ChangeLog.md
• The documentation has been updated, if necessary.

Please also shortly describe how you tested your change. Bonus points for added tests!

@dbaynard fixed #4390. Could you please confirm it?

[snoyberg]

I like this combinator 👍

[snoyberg]

It doesn't look like the updateOpts is really useful here, how about just set (globalOptsL...) (Just True)? I think it would be clearer that this is always being set.

[waddlaw]

Right!

[waddlaw]

Thank you for the review. updated!

[dbaynard]

Great job, @waddlaw! I'll review shortly. I thought I might have a little longer to find you the next issue to tackle 🤩

[dbaynard]

Generally good — just a few minor changes required.

It would be good to add some tests, too, perhaps modifying the integration test that checks this for stack clean from the fix to #4181 (although that test needs a little work, too).

[dbaynard]

As @qrilka noted, this could do with some documentation. We may want to change the WithDownloadCompiler enum as a result.

[dbaynard]

This would be better as two lines; respecting --no-install-ghc in bug fixes and the lack of GHC requirement merged with the stack clean line above, and moved to Other enhancements.

[qrilka]

I think that 2nd second sentence is misleading, the flag was respected at least in version 1.7.1 already:

qrilka@qdesktop ~/ws/h/stack-test-ghc-7.8.4 $ cat stack.yaml 
resolver: ghc-7.8.4
qrilka@qdesktop ~/ws/h/stack-test-ghc-7.8.4 $ stack --no-install-ghc clean
No compiler found, expected minor version match with ghc-7.8.4 (x86_64-tinfo6) (based on resolver setting in /home/qrilka/ws/h/stack-test-ghc-7.8.4/stack.yaml).
To install the correct GHC into /home/qrilka/.stack/programs/x86_64-linux/, try running "stack setup" or use the "--install-ghc" flag.
qrilka@qdesktop ~/ws/h/stack-test-ghc-7.8.4 $ stack --numeric-version
1.7.1

[waddlaw]

It's perfectly clear, thanks!!

[dbaynard]

Suggested change

	-- helper function for update some option
	-- helper function to update an option

It would be good to document what the Bool represents.

[waddlaw]

@dbaynard I was able to do it.--system-ghc flag is needed that resolve this issue.

$ stack --numeric-version
1.9.1

$ stack --no-install-ghc --skip-ghc-check --system-ghc dot --external
strict digraph deps {
"clean" [style=dashed];
{rank=max; "rts" [shape=box]; };
"base" -> "ghc-prim";
"base" -> "integer-gmp";
"base" -> "rts";
"clean" -> "base";
"ghc-prim" -> "rts";
"integer-gmp" -> "ghc-prim";
}

$ stack --no-install-ghc --skip-ghc-check dot --external
No compiler found...

msystem is Nothing if missing --system-ghc.

stack/src/Stack/Setup.hs

Lines 386 to 391 in 0b68efc

	msystem <-
	if soptsUseSystem sopts
	then do
	logDebug "Getting system compiler version"
	getSystemCompiler wc
	else return Nothing

stack/src/Stack/Setup.hs

Lines 395 to 399 in 0b68efc

	let canUseCompiler compilerVersion arch
	| soptsSkipGhcCheck sopts = True
	| otherwise = isWanted compilerVersion && arch == expectedArch
	isWanted = isWantedCompiler (soptsCompilerCheck sopts) (soptsWantedCompiler sopts)
	needLocal = not (any (uncurry canUseCompiler) msystem)

In other words, needLocal = not (any (uncurry canUseCompiler) Nothing).

ghci> :t any
any :: Foldable t => (a -> Bool) -> t a -> Bool

ghci> any (const True) Nothing
False

I think that we should just only enable these option in dot (and ls), or add new --skip-ensure-compiler flag. (This flag is like completely skip ensureCompiler function)

What do you think?

[qrilka]

I don't think enabling --system-ghc is a way to go here - it could be of a different and maybe incompatible version or there could be no system GHC at all. And the last one is the starting use case in #4390
Also I don't think that having a CLI flag --skip-ensure-compiler is a good idea - it's better to have that option internal-only and disable it only for a selected number of commands which should not fail without proper GHC available

[waddlaw]

Thanks @qrilka. I understand it.

My implemented checkDownloadCompiler is possibly wrong...?(always skipping ensureCompiler if --no-install-ghc specified)

checkDownloadCompiler _     SkipDownloadCompiler = return (Nothing, Nothing, False)
checkDownloadCompiler sopts WithDownloadCompiler
  | installIfMissing = ensureCompiler sopts
  | otherwise = return (Nothing, Nothing, False)
  where
    installIfMissing = soptsInstallIfMissing sopts

Hence, stack --no-install-ghc --skip-ghc-check dot --external works fine.

[qrilka]

I think it's wrong, yes, disabling GHC installation doesn't mean that compiler check should get skipped. For dot we should be able to go on without a compiler so SkipDownloadCompiler (or SkipCompilerCheck) should be OK for it. At least that's my understanding of these things.

[waddlaw]

All right. It was my mistake. I will fix asap!

[dbaynard]

There are some changes in progress (e.g. #4500) which may affect this.

[waddlaw]

@dbaynard
Sorry for the delay in work.I think I will work on it.
Is it better to work after # 4500 is merged?

[dbaynard]

It should be merged soon — but if you work on it before then just take into account those changes. I think the only changes left for that PR are in the test suite.

[snoyberg]

I'm looking into this right now as part of my work on cleaning up Stack.Runners. I'm confused as to why this is supposed to work at all. stack dot depends on knowing information about dependencies of global packages, which in turn requires having a locally installed GHC. It seems like the only way to make this work is to ignore global packages entirely, which this PR isn't doing.

[qrilka]

@snoyberg previously we didn't look into global packages and that's why I filed #4324

[snoyberg]

Thanks for clarifying @qrilka. I've opened up #4632, which will supersede this PR.