Run hs2hs on .hsc-files before loading

[unhammer]

Disable by setting haskell-process-do-hsc2hs to nil.

Tests only work interactively, TODO.

Fixes #1012.

[gracjan]

We would not want to push the burden on users with this variable due to cost described in https://github.com/haskell/haskell-mode/wiki/Customizable-variables

[unhammer]

We would not want to push the burden on users with this variable due to cost described in https://github.com/haskell/haskell-mode/wiki/Customizable-variables

OK; PR updated.

[gracjan]

Note that on our build server there is no hsc installed and in general we would love to not depend on hsc. Lets do this:

1. Add a variable that is hsc-exec-path (with more proper name).
2. In tests customize this variable to a shell script that pretends to be a hsc compiler.

So basically we mock hsc on shell level. What do you think?

[unhammer]

1 Add a variable that is hsc-exec-path (with more proper name).
2 In tests customize this variable to a shell script that pretends to
be a hsc compiler.

So basically we mock hsc on shell level. What do you think?

Sounds good to me. Do we customise test variables in haskell-hs2hs-tests.el, or is there a central file?

[gracjan]

In haskell-hs2hs-tests.el. Actually you could use dynamic binding so that customization is done per test.

[unhammer]

Added a fake script now; how does it look? I still get Wrong type argument: stringp, ("ghci" "-ferror-spans") as in #1012 (comment) both in my terminal and from travis though. Are there any other tests that load the REPL that I could look at?

[gracjan]

This should always use the hsc2hs.sh script. I do not want reports from people that accidentally have something strange in their paths and therefore tests accidentally pass or fail.

[unhammer]

done

[gracjan]

This Wrong type argument: stringp, ("ghci" "-ferror-spans") indicates a couple of things:

1. Something in there should compose cmd line invocation even given list of strings. See Improve error handling of exec of external programs if they are absent and handle list of params #975.
2. On Travis we do not want ghci so there is a need for another awk script fake-ghci.sh just like the one you did for hsc2hs.
3. In general when executables aren't found the information should be much more user friendly.

[gracjan]

And also note that this whole external program execution testing is tracked by issue #1206.

[gracjan]

Nice!

[gracjan]

How about inlining the contents of hsc2hs.sh in a function that will create this script on the fly, add executable bit and setup the path, I mean:

     (with-script-path ' haskell-process-path-hsc2hs
      "#!/bin/awk
      /END/ { the script here }"
      (body-of-function))

So this will create a new file with contents given, add executable bit, and set the given variable to the path of what was created. What do you think?

[unhammer]

That'd work, but the macro can't clean up the file because it's running async now. Does
https://github.com/unhammer/haskell-mode/compare/hsc2hs-1012...unhammer:hsc2hs-1012-tempscript?expand=1 seem OK?

[gracjan]

Yes.

No cleanup is okay, we might fix that later as low priority.

[gracjan]

You need to create a script that plays the role of GHCi.

[unhammer]

I haven't been able to get that to work; the process seems to just die before it sends any output at all.

Does haskell-repl send any signals or such before interaction happens? Should the output be different from running ghci on the command line?

[gracjan]

I do not think there are any signals involved. If there is an async communication isn't that the tests ends up before fake-ghc?

[unhammer]

I was checking (process-status (get-buffer-process "*haskell*")), but now I see there is never any process associated with the REPL buffer; doing (process-status (cdr (assoc 'inferior-process (haskell-process)))) actually checks the status, so it is running after all.

I'm just not getting the output in the REPL. I see the output in the haskell-process-log of the fake-ghci if I, instead of (insert ":t unNUMBERS rand_max") (haskell-interactive-handle-expr) do (haskell-process-send-string (haskell-session-process haskell-session) ":t unNUMBERS rand_max") (is that just for "internal" communication?), but it doesn't end up in the REPL.

[gracjan]

Interesting. Communication is somehow preprocessed, maybe something is eating your output?

[gracjan]

Ping? Anything happening here?

I find this idea really good, so would be lovely to see this pushed forward!

[unhammer]

Hey; I never figured out how to fix this :( (also, I haven't actually had use for hsc2hs in some months months).

I'd love to get it working though. Could you explain what happen (or is supposed to happen) between the haskell-process-log and the REPL? I see haskell-mode at least puts the lambda character in the prompt, presumably some other things happen there as well. Since the log shows the expected output, I'm guessing whatever is parsing my fake-ghci is not seeing what it expects.

[gracjan]

Well, I do know as much as you do at this point.

A fake-ghci would be great to test other things in haskell-interactive-mode.

[gracjan]

Anyway I plan to steal the part with awk, make it more general and use it for some other tests.

[unhammer]

Cool; it'd be great if something useful comes of this :)

[gracjan]

I've remade the testing parts in #1331, no awk, just pure elisp :)

If you happen to need hsc again then please use my testing functionality.

Thanks for inspiration!