cabal v2-repl scriptfile.hs

[phadej]

cabal v2-repl scriptfile.hs outside project to load a scriptfile.hs into repl (GHCi).

---

Suppose I have a script file like:

{- cabal:
build-depends: base, splitmix
-}

import Data.List (unfoldr)
import Data.Word (Word64)
import System.Random.SplitMix (nextWord64, mkSMGen)

numbers :: [Word64]
numbers = unfoldr (Just . nextWord64) (mkSMGen 42)

main :: IO ()
main = print $ take 10 numbers

cabal v2-run -v0 scriptfile.hs works perfectly, however it takes quite a lot of time, 10s on my machine!

% time cabal v2-run -v0 ExRun.hs
[1275548033995301424,10417309031967933079,2112719111588962399,2726445820918627087,4609004260007739600,11074269206337254866,9733109294540808081,6450629194222476823,8084222243108663947,16253064923614931039]
cabal v2-run -v0 ExRun.hs  10,70s user 1,01s system 99% cpu 11,724 total

Slowness is one issue, after all there's compiling involved. It however makes development of scriptfiles extremely painful.

Yet, if the following worked, the development would be a lot nicer (repl ftw):

% cabal v2-repl ExRun.hs
Resolving dependencies...
Build profile: -w ghc-8.6.5 -O1
In order, the following will be built (use -v for more details):
 - fake-package-0 (exe:script) (first run)
Configuring executable 'script' for fake-package-0..

...

GHCi, version 8.6.5: http://www.haskell.org/ghc/  :? for help
[1 of 1] Compiling Main             ( ExRun.hs, interpreted )
Ok, one module loaded.
*Main> 

*Main> main
[1275548033995301424,10417309031967933079,2112719111588962399,2726445820918627087,4609004260007739600,11074269206337254866,9733109294540808081,6450629194222476823,8084222243108663947,16253064923614931039]

worked, it would be super great.

---

For now, I try to cabal install --lib the dependencies, and use raw ghci.

[mpickering]

I'm interested in this feature. Could you give some pointers about how hard it would be to implement? (cc @fendor)

[fendor]

I intend to work on this.
The implementation of the scriptfile feature seems to be mainly in CmdInstall. It sets up a fake project by essentially writing a *.cabal file to a temporary location and then building this fake project.

First, we should move implementation details from CmdInstall to its own module, proposing: CmdScript.hs.
The implementation itself seems trivial (we will see), the only difficulty I anticipate is to detect changes in the meta-data section of the script on reload (e.g. :r in the ghci session).

[phadej]

The fake package stuff is better put into Distribution.Client.FakePackage something like that. CmdScript would be confusing as there could be cabal script functionality (but it's part of run).

[phadej]

It's worth to make clear to one self how cabal v2-build, cabal v2-run and cabal v2-repl behave outside and inside project. Perfectly these won't have different behaviors / cli interfaces, and I think the easiest way to get there is to abstract common parts more heavily.

[fendor]

so, the first implementation was actually rather trivial.
The current mechanism is to copy the file to some /tmp/cabal-repl. directory and generating a very basic fake-package.cabal. So, we can just open a repl session in that project context. (Although, admittedly, it doesnt make sense to set the cwd to /tmp/cabal-repl. but that is a task for later).
The hard problem is to use :r in the repl session. I would expect that this re-compiles changes, such as source changes and changes to the meta-data.
However, since the source file is copied, neither of these changes can be detected. I had the following ideas:

• Symlink Script.hs to /tmp/cabal-repl.*/Main.hs
  
  
  
  • doesnt reliably work on windows
  
  
  • propagates updates the source file
  
  
  • doesnt work to update meta-data
  
  
• Only write cabal.project and fake-package.cabal to /tmp/cabal-repl.*/
  Now use semantically the invocation cabal repl Script.hs --project-file /tmp/cabal-repl.*/cabal.project
  
  
  
  • propagates updates from the source file
  
  
  • still doesnt work for updating meta-data
  
  

Are there more options? Can we hook somehow into the reload command of the repl-session to automatically get notified about changes to Script.hs ?
Btw, stack manages to do both, on :r changes to the source file and meta-data are reloaded appropriately.

---

It's worth to make clear to one self how cabal v2-build, cabal v2-run and cabal v2-repl behave outside and inside project.

they are quite similar indeed!

[phadej]

It is ok to not support updating on meta-data change. cabal repl doesn't update anything when you update .cabal file either. Changing meta-data might change everything in GHCi setup. In other word: let's not try hook anything to :r command. I say that's out-of-scope.

One way to approach this is maybe to make cabal repl Script.hs to load current project and the Script.hs first. Then no-project case (this issue) would be cabal repl /absolute/path/to/Script.hs in the fake project context.

[fendor]

So, workflow is roughly as follows cabal repl Script.hs:

• Create a fake-package.cabal with the build-depends and main-is: Main.hs.
• create ghci-script with the content:
  :cd <cwd> :add <cws>/Script.hs
• Loading this project. This will compile Main.hs as well as <cwd>/Script.hs and add Script.hs to the repl and set the working dir correctly. Changes to Script.hs are detected on :r.

What I dont like about this is that two files are loaded, Main.hs and <cwd>/Script.hs. It seems I cant set main-is: <cwd>/Script.hs, according to the docs, and I admit that I prefer it that way.
However, then we have to set some main-is and load this file on cabal repl Script.hs.

A work-around I have been thinking about:

• Create library stanza instead of executable in fake-package.cabal.
  
  
  
  • The meta-data part is parsed by executableGrammar, so that does not really work
  
  
  • Using a different stanza when opening a repl for the script is counter-intuitive since there is a different support of fields.
  
  

Any other good way of handling this?