Ihor Radchenko <yantar92@posteo.net> writes:

> We can try the attached patch. Yet another edge case in comint, it
> appears.

I investigated further and now applied a set of patches that improves
prompt filtering in org-comint.

https://git.savannah.gnu.org/cgit/emacs/org-mode.git/commit/?id=a8a516ba330fe8c435334030ffbe371b8c80c877
https://git.savannah.gnu.org/cgit/emacs/org-mode.git/commit/?id=866ed1a3c5c37cad243085f9a8fa904970e4d614

Further, I tried to incorporate Pareto's suggestion about secondary
prompt and cleaned up the original GHCi tests.

With the attached set of patches (first one is Bruno's original patch),
I get all the tests reliably passing, except the tests left as
:expected-result failed.

The remaining tests fall into two categories:
1. Tests trying to test :results value vs. :results output in sessions.
2. Tests trying to test for multiple sessions not interfering each
   other.

The first category of tests fails because `org-babel-interpret-haskell'
simply cannot distinguish between :results value and :results output.
Someone familiar with GHC should handle this problem. For example, by
wrapping :results output code in such a way that output is discarded in
the GHCi comint buffer.

The second category of tests fails because haskell-mode uses a fixed
"*haskell*" comint buffer name. As I suggested in another message, a
viable approach could be renaming the original "*haskell*" buffer to
something else in `org-babel-haskell-initiate-session'. Again, someone
more familiar with haskell-mode should judge if such option is truly
viable.