'cabal repl' with stack traces

[emlautarom1]

I've been looking around for a set of flags/configurations that would allow me to load my project into GHCi and have access to stack traces. So far, I found this blog post that explains how to access stack traces from plain GHCi (not from cabal repl). I also found this in the official docs - which is practically a summary of the blog post.

What I've tried so far with no success:

• Add -fexternal-interpreter -prof to the ghc-options section in my .cabal file
• Run cabal repl --ghc-options="-fexternal-interpreter -prof". I got a warning from GHCi, but no stack traces.
• Run cabal repl -p. This required installing the ghc-prof library (I'm using Fedora).
• Run cabal exec ghci -- -fexternal-interpreter -prof. I got some errors related to missing files in the new-dist folder. Then, I recompiled with cabal build -p.

My project has several targets (a lib, a test suite and several executables). What I want is to load the library (the default when running cabal repl) and use Debug.traceStack. I do not care about getting stack traces from third party dependencies, I just care about what my code is doing.

Right now I'm using HasCallStack as a workaround, but the traces are extremely short (a single function) and it's very inconvenient to add the constraint on every function.

[fgaz]

What were the results of the last two bullet points? -p should work iirc

[emlautarom1]

Minimal working example: run cabal init --lib. Edit MyLib.hs with the following code:

module MyLib (someFunc) where

import qualified Debug.Trace as Debug

f :: Int -> Int
f x = Debug.traceStack "> from f" (x + 1)

someFunc :: IO ()
someFunc = do
  let x = 5
  let res = f x
  putStrLn "result"
  print res
  putStrLn "done"

Start REPL using cabal repl -p. Evaluate someFunc. Output is:

*MyLib> someFunc 
result
> from f
6
done

I expect to get a stack trace, yet I get none. It behaves like Debug.trace.

[emlautarom1]

I'm using cabal 3.0 by the way (cabal-install version 3.0.0.0; compiled using version 3.0.1.0 of the Cabal library)

[fgaz]

cabal build -p
cabal exec ghci -- -fexternal-interpreter -prof

works for me. I'm using master, but I'd espect 3.6 to work too, please say so if not (3.0 is quite old by the way!)

Still, I think profiling flags should be passed to ghci in cabal repl as well. The one above is just a workaround.

[emlautarom1]

Indeed, that works. I need to load the library once in GHCi though (:l MyLib).
Sadly this does not work on my main project which uses several third party dependencies. I get errors related to shared objects (after running cabal exec ghci -- -fexternal-interpreter -prof):

GHCi, version 8.8.4: https://www.haskell.org/ghc/  :? for help
Loaded package environment from /home/emlautarom1/Development/Haskell/duckling/dist-newstyle/tmp/environment.-176033/.ghc.environment.x86_64-linux-8.8.4
Loaded package environment from /home/emlautarom1/Development/Haskell/duckling/dist-newstyle/tmp/environment.-176033/.ghc.environment.x86_64-linux-8.8.4
<command line>: libHSextensible-exceptions-0.1.1.4-66ee67394de71dfcdfc753af01d17a538e5fbe79e155b15ebbc55afc2a4fffe8.so: cannot open shared object file: No such file or directory

I found that I could fix this issue by installing ghc-extensible-exceptions but then there's another issue:

Resolving dependencies...
GHCi, version 8.8.4: https://www.haskell.org/ghc/  :? for help
Loaded package environment from /home/emlautarom1/Development/Haskell/duckling/dist-newstyle/tmp/environment.-176275/.ghc.environment.x86_64-linux-8.8.4
Loaded package environment from /home/emlautarom1/Development/Haskell/duckling/dist-newstyle/tmp/environment.-176275/.ghc.environment.x86_64-linux-8.8.4
<command line>: libHStimezone-series-0.1.9-1ee495d5ea91c01050b65c31aa846fca19aa17022b9c45c7ed21d61bb40e17d8.so: cannot open shared object file: No such file or directory

So, again, I need to install ghc-timezone-series. I guess that I would need to do that for every dependency, which seems totally unreasonable. Could be the fact that I installed GHC through my package manager (dnf) and not stack or ghcup the cause of those missing shared objects?

---

Edit: actually, ghc-timezone-series is not available through dnf, so I can't even try installing the package.

[fgaz]

Is that with cabal-install 3.6? Are you using plain commands (build, repl, exec...) or version-prefixed commands (v1-build, v1-repl, v1-exec... / v2-build, v2-repl, v2-exec...)

[emlautarom1]

I'm using cabal 3.0, plain commands which default to v2 versions. I haven't tried it yet with newer versions since they're not available through dnf. I could try using the latest version just to ensure that it works as expected now.

[Mikolaj]

With some luck cabal install cabal-install should bring you new cabal, so you don't have to use any special tool for that.