[antlr-interest] executing actions while recovering from syntax errors

[Terence Parr]

Hi,

Following Josef Grosch's nice error recovery work in CoCo, ANTLR v3 
goes into an "error recovery mode" upon syntax error.  It invokes 
reportError() and then recover() (with an exact FOLLOW set).  After 
recovery, it attempts to continue parsing.  Until a token is 
successfully matched, however, ANTLR is still in "recovery mode" 
meaning that no more messages will be spit out.

So, should I be executing actions during recovery mode?  Seems to me 
that until the next we match a token properly, actions may not be 
executing in the environment they expect.  For example,

decl : type ID '=' expr {action2} ';' ;
....
expr : ... {action1} ... ;

If an error occurs in expr, action1 may not be executed.  After 
recovery (consuming tokens to resync), expr returns to the invoker: 
decl.  At this point, action2 may fail (with an exception or just 
produce bad results) because action1 didn't execute.  ANTLR is back out 
of recovery mode when it sees the ';' probably.  At that point, it may 
be ok to execute actions.

'course once a syntax error occurs, one could argue no action is "safe".

Anyway, putting a guard around every action looks gross and takes 
time/space (oh, and I guess is no guarantee it's safe):

if ( !recoveryMode ) { action }

Anybody wanna comment?  For now, the actions are unguarded.

Ter
--
CS Professor & Grad Director, University of San Francisco
Creator, ANTLR Parser Generator, http://www.antlr.org
Cofounder, http://www.jguru.com