Guard haskell-process-do-try-info from non-strings

[deviant-logic]

haskell-process-do-try-info will break process interaction if it isn't passed
a string. This can happen when using haskell-mode-contextual-space and
typing a space after ::. At least, it happened to me when I did it.

[gracjan]

@deviant-logic: In elisp the usual protocol agreement is that functions are not called with arguments falling outside of their domain.

So the proper solution is to fix haskell-mode-contextual-space or even better is to fix the source of this problem that is described in issue #334.

So I would gladly merge changes to either haskell-mode-contextual-space or haskell-ident-at-point.

Can you create such a pull request?

[deviant-logic]

@gracjan: I can fix haskell-mode-contextual-space tomorrow; I'm up way
past my bedtime at the moment. Having haskell-process-do-try-info be
defensive may be justified here regardless, though, as its failure-mode
breaks process interaction. I'm ambivalent. Since
haskell-ident-at-point is documented to return nil when it can't find
an identifier, mucking with that function strikes me as inappropriate here.

On Sat, Apr 25, 2015 at 12:49 AM gracjan notifications@github.com wrote:

@deviant-logic https://github.com/deviant-logic: In elisp the usual
protocol agreement is that functions are not called with arguments
falling outside of their domain.

So the proper solution is to fix haskell-mode-contextual-space or even
better is to fix the source of this problem that is described in issue
#334 #334.

So I would gladly merge changes to either haskell-mode-contextual-space
or haskell-ident-at-point.

Can you create such a pull request?

—
Reply to this email directly or view it on GitHub
#601 (comment).

[gracjan]

We should try to follow the usual practice of emacs lisp and that is callers are responsible for providing argument of correct type. Please fix haskell-mode-contextual-space when you have time. And please look at haskell-ident-at-point as we had reports that it does not follow its own documentation.

[antalsz]

@gracjan: haskell-ident-at-point now follows its own documentation (it returns nil when there isn't an identifier), but haskell-ident-pos-at-point does not (it still returns a pair of the form (n . n)) – I can fix that. Perversely, however, there seems to be plenty of code which assumes that haskell-ident-at-point does not follow its own documentation and returns "" (including possibly the root cause of the haskell-process-do-try-info bug).

I can try to fix this, but I'm not sure where I should put that commit – a new PR, here, on #602?

Also, a +1 for fixing the crash bug in haskell-process-do-try-{info,type} regardless (either @deviant-logic's fix or mine), because it renders the GHCi buffer completely unusable unless you kill the session. (But yes, fixing the root cause is better.)

[gracjan]

I would prefer another pull request that fixes everything about mishandling of haskell-ident-at-point and haskell-ident-pos-at-point.

After that is done we will look again on #601 and #602.

[gracjan]

Also note that haskell-ident-at-point has a testsuite and tests for haskell-ident-pos-at-point should also get a mention in the test suite.

[antalsz]

@gracjan, @deviant-logic: I submitted pull request #603, which fixes the behavior around haskell-ident-at-point, thereby fixing haskell-contextual-space and the problem here.

Thinking again about whether haskell-do-try-info ought to be defensive: perhaps the real issue is that an error inside the :go callback of make-haskell-command completely fouls up process interaction permanently. Maybe that's what really ought to be fixed?

[gracjan]

If I understand correctly this was subsumed by #603, right?

[gracjan]

@antalsz: About :go being unprotected. This is something we should definitely look into. Chris rolled about home grown async/await kind of functionality probably because there was no such thing available when he did it. I guess that there should be something like that now and we should find something working out-of-the-box.