ContentAssist / CompletionScanner running into deadlock

[srikanth-sankaran]

What it does

• Ensure that attempts to nudge recovery along don't stuck reducing empty statements

• Fixes ContentAssist / CompletionScanner running into deadlock #1100

Author checklist

• I have thoroughly tested my changes
• The change is following the coding conventions
• I have signed the Eclipse Contributor Agreement (ECA)

[srikanth-sankaran]

In the method org.eclipse.jdt.internal.codeassist.impl.AssistParser.fallBackToSpringForward(Statement), there is a comment that reads:

// OK, next token is no good to resume "in place", attempt some local repair. FIXME: need to make sure we don't get stuck keep reducing empty statements !! - this was added by me in 2014 November - my last month in inning no. 1, it has taken a long while to get to inning no. 2 and to the FIXME :-(

Anyways, code completion is an extremely complex area nothing short of dark magic as it stands - there is so little good quality documentation on it, there is so much global state that is scattered in bits and pieces each dealing with some corner case or other. Each successive maintainer has tried (and failed IMHO) to unearth the design and has ended up adding his own complex layers on top.

This is not an intractable area in truth, but in practice very much so because most committers have no time to grasp the full picture, but do as well as the blind men in the fable with elephants.

The present ticket is even more tricky - CompletionParser ends up seeing lambdas where there are none! Things go downhill pretty quickly from there. I have added the smarts to detect where recovery is not progressing at all despite our attempts to nudge it along - by tracking the parser's and scanner's state to deduce deadlock and break free.

This would prevent eclipse from hanging. (There are no useful completion proposals at the site)