search

isaacs / github

Just a place to track issues and feature requests that I have for github
2.2k stars 129 forks source link

Parse commit messages intelligently #357

Open RichardLitt opened 9 years ago

RichardLitt commented 9 years ago

For instance, the commit:

Fixes the issue in #72

ought to be parsed, and it should be noted that the issue number is in a prepositional phrase and that Fixes applies to #72. An easy way to do this is to use a parser to check the object of the commit sentence, and to see if there is a clause between fixes and the issue number.

cirosantilli commented 9 years ago

I feel this is too much trouble for what it is worth for GitHub. And wouldn't it lead to more false positives?

RichardLitt commented 9 years ago

I want to live in a world where I can do this. I'm not really mindful of whether or not GitHub has the resources to improve their NLP. It would only lead to false positives if trained poorly.

cirosantilli commented 9 years ago

If you have some knowledge in the domain, I recommend that you point to library that does it well, possibly with sample code, specially if it is in Ruby. That would probably greatly increase the chance of it happening.

RichardLitt commented 9 years ago

I don't know the NLP ecosystem in Ruby. I would look at NLTK in Python or Natural in Javascript, though.

rlidwka commented 9 years ago

"This might fix an issue in #xxx if yyy under zzz conditions." I'd hate it if this message will close an issue.

The less intelligent github is about it, the better it is for users imho. 'cause false positives.