Repository navigation
added parsing for PostgreSQL operations - #267
Conversation
Pull Request Test Coverage Report for Build 278078999
💛 - Coveralls |
|
@nickolay can you please take a look? |
|
This seems good. Could you fix the conflicts with the recent merge, and use the |
5e10970 to
0df8011
Compare
|
@nickolay can you have a look? |
| chars.next(); // consume the '|' | ||
| match chars.peek() { | ||
| Some('|') => self.consume_and_return(chars, Token::StringConcat), | ||
| Some('/') if dialect_of!(self is PostgreSqlDialect) => { |
There was a problem hiding this comment.
@nickolay, am I right that this is not going to work with custom dialect as you suggest here #243 (comment)?
There was a problem hiding this comment.
I'm sorry, but I don't follow. In the comment you mention I said I didn't think that simply allowing $ to start an identifier matched what the PosgreSQL documentation said about this.
In this PR you make |/ conditionally parse as SquareRoot, which makes sense to me.
There was a problem hiding this comment.
I have custom Dialect https://github.com/alex-dukhno/database/blob/master/src/parser/src/lib.rs#L221-L231, is it true that it wouldn't parse |/ into SquareRoot because of if dialect_of!(self is PostgreSqlDialect)?
There was a problem hiding this comment.
Ah, I see now. Yes, I assumed |/ is something very Postgres-specific, so limiting it to PostgreSqlDialect (and GenericDialect, which I didn't mention explicitly) seemed to make sense.
In the older comment you're referring to I described a stop-gap measure that would be used until someone figured out the proper way to deal with $... in the parser.
We could make the new operators parse unconditionally for now if that makes your life easier, but without a clear use-case (and accompanying testcase) this will likely break in the future.
There was a problem hiding this comment.
I think it's better to have dialect_of checks and have a solution for dealing with $.... Having my knowledge of crate I'd suggest to add Variable(Ident) or something similar to ast::Expr enum. It has to be done in a separate PR of course. WDYT?
There was a problem hiding this comment.
I have @-mentioned you in #265, where this discussion can continue.
The hard part is describing what specifically we're trying to implement and why. Your suggestion, for example, solves the problem of representing $ (but not another leader) followed by an identifier (quoted or not; consisting of characters from an unspecified set) anywhere an arbitrary expression is allowed (but not in more restricted contexts, like OFFSET __ ROWS). I'm not sure if this matches Postgres or the "common denominator" of the usual dialects; answering that requires some research. From the examples I've seen so far, a new Value variant with {leader, ident} fields might be more appropriate.
nickolay
left a comment
There was a problem hiding this comment.
I'm sorry for taking so long!
| chars.next(); // consume the '|' | ||
| match chars.peek() { | ||
| Some('|') => self.consume_and_return(chars, Token::StringConcat), | ||
| Some('/') if dialect_of!(self is PostgreSqlDialect) => { |
There was a problem hiding this comment.
I'm sorry, but I don't follow. In the comment you mention I said I didn't think that simply allowing $ to start an identifier matched what the PosgreSQL documentation said about this.
In this PR you make |/ conditionally parse as SquareRoot, which makes sense to me.
| Token::CubeRoot => Ok(Expr::UnaryOp { | ||
| op: UnaryOperator::PGCbrt, | ||
| expr: Box::new(self.parse_subexpr(0)?), |
There was a problem hiding this comment.
This is probably not right, as this parses WHERE ||/27 = 3 as ||/ ( 27=3 ), which is not a valid expression. I don't see precedence mentioned in the documentation you linked to, and I think a better guess would be to default all the unary operators to the PLUS_MINUS_PREC (moving the new ops to the tok @ case below).
094bad4 to
1f2b8d3
Compare
There was a problem hiding this comment.
I'm ready to merge this, waiting for your decision on removing the dialect-specific parsing.
| } else if Token::DoubleColon == tok { | ||
| self.parse_pg_cast(expr) | ||
| } else if Token::ExclamationMark == tok { | ||
| // PostgreSQL factorial operation | ||
| Ok(Expr::UnaryOp { |
There was a problem hiding this comment.
Ideally we'd consider how postfix factorial interacted with other operators (the postfix ::type, another factorial, and prefix operators), but I don't think this has to block merging this PR.
|
Excellent, thank you! |
Added support to parse mathematical operations for PostgreSQL
See table 9.4