sql2 parser bug with operator precedence

[dbu]

see phpcr/phpcr#35 for discussion

parenthesis are parsed by the sql2 parser. however, operator precedence is ignored: a or b and c gets parsed as if it was (a or b) and c. the parser should respect precedence of operators.

i suggest starting with a couple of phpcr-api-tests that test operator precendence and parenthesis (and do the assertions really based on what would be correct and not on what you get from the current implementation ;-) )

fixed
the qom to sql2 generator adds parenthesis everywhere to avoid potential problems.

[chregu]

Just adding parenthesis around every and and or constraint seems ok to me. I'd prefer too many parenthesis to none at all ;)
At least for the quick fix and if we starting to recommend the QOM, then this is even more important

[dbu]

#16 fixes the sql2 generator. the parser is still broken.

[dbu]

@igorw said "dijkstra's shunting yard algorithm."

[fabpot]

I would recommend using the precedence climbing algorithm. That's what I use in Twig (https://github.com/fabpot/Twig/blob/master/lib/Twig/ExpressionParser.php) and it is very easy to implement and it allows to reduce the number of nested calls to a minimum (which is always nice when you are using XDebug). See http://en.wikipedia.org/wiki/Operator-precedence_parser for more information.

The http://www.engr.mun.ca/~theo/Misc/exp_parsing.htm page is also worth a read as it explain the benefits of this algorithm versus the shunting-yard algorithm.