Discussion about error recovery

[gvanrossum]

Maybe we can continue the discussion (started here of my error recovery code here.

I could think of several different topics:

• How the recovery actually works
• How it compares to the paper
• How I am validating it (see https://github.com/we-like-parsers/pegen/blob/master/corpus.py)
• What I'd like to do next

Here's an incomplete TODO list:

• Switch corpus.py to using a sqlite3 database, so we can store past results and do queries (on input and output)
• Find more corpora -- the corpus I found is skewed heavily towards copy/paste errors into StackOverflow, unfortunately
• Come up with more special cases for make_improved_syntax_error