Re: Pluggable Parser

Rajeev rastogi <rajeev(dot)rastogi(at)huawei(dot)com> writes:
> What is Pluggable Parser:
> It is an option for users to select a different kind of parser to evaluate PostgreSQL for their business logic without much manual effort.

> Why do we need?
> As of now migration from other databases to PostgreSQL requires manual effort of translating SQL & PL/SQL to PostgreSQL equivalent SQL queries. Because of this manual effort in converting scripts, migration to PostgreSQL considered to be very slow and sometime de-motivating also. So if we allow to plug different database syntaxes with PostgreSQL, then it will be one of the strong motivating result for many DBAs to try PostgreSQL.

While I don't have any strong reason to object to putting a hook where you
suggest, I think that the above represents an enormous oversale of the
benefits, which in actual fact are likely to be near zero. Replacing
gram.y as you suggest will not allow more than the most trivial, cosmetic
grammar changes, because you'd still have to produce the same raw parse
trees as before.

Alternatively you could consider replacing both raw_parser() and
parse_analyze(), but then you're talking about maintaining a duplicate
copy of just about the whole of src/backend/parser/, which doesn't sound
terribly practical; it certainly would be unpleasant to maintain such a
thing across multiple PG releases. And there will still be pretty strong
constraints on whether you could implement say a MySQL-workalike, because
you're still having to work with the Postgres execution engine.

In short, this proposal sounds a lot like a solution looking for a
problem. I think it would be useful for you to pick a small number of
concrete alternative-grammar problems and think about how you could
provide hooks that would allow solving those issues without duplicating
90% of src/backend/parser/ first.

Here is an example of such a concrete problem: Oracle, I believe, uses
parentheses () rather than brackets [] for array subscripting expressions.
How would you allow that notation in Postgres? Your original proposal
cannot do it because it's not a problem of the raw grammar, but rather
one for parse analysis --- you have to disambiguate array references
from function calls at the time of doing semantic analysis.

Another example you might consider is Oracle-style outer join notation,
which again will require significant surgery in parse analysis, not
just modifying the raw grammar.

regards, tom lane