Gregory Stark <stark@enterprisedb.com> writes:
> So instead of substituting them as the tokens are lexed, instead suck in the
> tokens, run the parser -- which we currently do anyways just to check the
> syntax -- then walk the tree looking for ColumnRefs where the name matches a
> variable name. Then keep around that parse tree instead of just the series of
> lex tokens to later call analyze on and execute.
Hmm. That sounds cool but I think it actually has a pretty substantial
disadvantage --- there is then no easily user-readable representation of
the query that shows *which* occurrences got substituted. With the
textual replacement method you have a string you can look at, though
it's true that we don't always show it to the user if we are not aware
there's a problem.
Again, I'm trying to look at the big picture of both syntactic and
semantic errors. If we solve only the syntactic end of it I think we'd
actually be worse off, because then users would be even more lost when
they hit a semantic error (unwanted substitution).
regards, tom lane
