Hinterland should only hint for nontrivial prefixes

[dmyersturnbull]

Steps to reproduce (simplest example):

1. In an empty line in a Python cell, press # to start a comment.
2. Press enter to end the line.

Actual result: The line now reads #%%! (the first completion).

Expected result: The line should be an empty comment (just #).

The above example isn't especially annoying, but in a day of using Hinterland, I've repeatedly had to trim %%! from meaningful lines of code, not just comments.

[jcb91]

Yes, this was a problem I ran into when writing it with the colon :. It's made a little more tricky to get right in general because of not knowing which language kernel may be in use. As a result, I think perhaps it makes sense to have a user-configurable set of characters after which hinting is not triggered, with the default including the #.

I've repeatedly had to trim %%! from meaningful lines of code, not just comments.

Is this a result of other problematic characters, or something worth addressing more generally?

[jcb91]

@dmyersturnbull any suggestion for what constitutes a non-trivial prefix?

[dmyersturnbull]

@jcb91 Good question. Perhaps trivial prefixes are just ^[^A-Za-z\!]$ with the ! just to catch external commands. That might be too conservative for a definition of trivial.

[jcb91]

Hmm, I hadn't thought about the external command prefix. I'm not sure I follow your regex, given the ^[^A-Za-z, you want to consider trivial anything beginning with a non-alphabetic character - wouldn't this include the external-command ! already? Unless you mean you think external-command lines should be considered non-trivial, but for me they don't give useful completion suggestions, so that seems unlikely? Also, does this test apply to just the character preceding the cursor (seems like it would be enough?), or to the entirety of the line before the cursor? It's possible also that people will want to use unicode characters for variables etc, which would put a hole in this regex, but I can't think of a (simple) better one right now, other than listing every character with a special definition in python syntax. Anyway, this is getting complicated, so I think it would be best to make this a configurable option, since it will also vary between kernel languages...

[jcb91]

Actually, it might be worth using the Completer.reinvoke_re at notebook/static/notebook/js/completer.js#L107 as a start point, since it seems from the comment that it may be adjusted to suit the kernel.

Probably worth also adding user-configurable positive & negative include/excludes (e.g. I don't like completion on colons, even though they are included in Completer.reinvoke_re)

[jcb91]

So, I've made a stab at configurable regexes, with the default as Completer.reinvoke_re, which by default is /[%0-9a-z._/\\:~-]/i, so should hopefully get rid of your problems with #. If you could give the PR a try, and let me know whether it's helpful, that'd be great

[jfbercher]

@jcb91 It seems very nice, and the way to go. I will give a try this afternoon.

[jcb91]

Assuming this has been fixed by #815 and #819. Please reopen if not!