Bruno Barbier <brubar.cs@gmail.com> writes:

> Ihor Radchenko <yantar92@posteo.net> writes:
>
>> From a first look, random failures appear to be related to session
>> initialization.
>
> My function 'test-ob-haskell-ghci' should protect against that; it
> ensures the "*haskell*" buffer is always new.  From what I understand,
> this is an input buffering problem: inputs are not full lines.

I see.

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