Snowflake diverges from the standard and from most of the other
implementations by allowing extra parentheses not only around a join,
but around lone table names (e.g. `FROM (mytable [AS alias])`) and
around derived tables (e.g. `FROM ((SELECT ...) [AS alias])`) as well.
Initially this was implemented in https://github.com/ballista-compute/sqlparser-rs/issues/154
by (ab)using `TableFactor::NestedJoin` to represent anything nested in
extra set of parens.
Afterwards we learned in https://github.com/ballista-compute/sqlparser-rs/issues/223
that in cases of such extraneous nesting Snowflake allows specifying the
alias both inside and outside parens, but not both - consider:
FROM (table_factor AS inner_alias) AS outer_alias
We've considered implementing this by changing `TableFactor::NestedJoin`
to a `TableFactor::Nested { inner: TableWithJoins, alias:
Option<TableAlias> }`, but that seemed too generic, as no known dialect
supports duplicate aliases, as shown above, nor naming nested joins
`(foo NATURAL JOIN bar) alias`. So we decided on making a smaller change
(with no modifications to the AST), that is also more appropriate to the
contributors to the Snowflake dialect:
1) Revert #154 by rejecting `FROM (table or derived table)` in most dialects.
2) For `dialect_of!(self is SnowflakeDialect | GenericDialect)` parse
and strip the extraneous parentheses, e.g.
`(mytable) AS alias` -> `(mytable AS alias)`
Co-authored-by: Eyal Leshem <eyal@satoricyber.com>
Since PR https://github.com/ballista-compute/sqlparser-rs/pull/93
`parse_column_def` parses a set of column options in a loop, e.g. given:
```
_______ column_def _______
CREATE TABLE foo (bar INT NOT NULL DEFAULT 1, )
-------- ---------
option 1 option 2
````
it parses column options until it encounters one of the delimiter tokens
First when we only supported `CREATE TABLE`, the set of delimiters that
stopped the parsing used to be `Token::Comma | Token::RParen`.
Then we added support for `ALTER TABLE ADD COLUMN <column_def>`. Turns
out the parser started to bail if the statement ended with a semicolon,
while attempting to parse the semicolon as a column option, as we forgot
to add it to the set of delimiter tokens.
This was recently fixed in https://github.com/ballista-compute/sqlparser-rs/pull/246
by including Token::SemiColon to the list, but it felt wrong to have
to update this list, and to have a common list of delimiters for two
different contexts (CREATE TABLE with parens vs ALTER TABLE ADD COLUMN
without parens).
Also our current approach cannot handle multiple statements NOT
separated by a semicolon, as is common in MS SQL DDL. We don't
explicitly support it in `parse_statements`, but that's a use-case
like to keep in mind nevertheless.
* Change `Parser { ... }` to store the dialect used:
`Parser<'a> { ... dialect: &'a dyn Dialect }`
Thanks to @c7hm4r for the initial version of this submitted as
part of https://github.com/ballista-compute/sqlparser-rs/pull/170
* Introduce `dialect_of!(parser is SQLiteDialect | GenericDialect)` helper
to branch on the dialect's type
* Use the new functionality to make `AUTO_INCREMENT` and `AUTOINCREMENT`
parsing dialect-dependent.
Co-authored-by: Christoph Müller <pmzqxfmn@runbox.com>
Co-authored-by: Nickolay Ponomarev <asqueella@gmail.com>
In MySQL it's AUTO_INCREMENT
(see https://dev.mysql.com/doc/refman/8.0/en/create-table.html)
and in SQLite it's AUTOINCREMENT.
We use `ColumnOption::DialectSpecific(Vec<Token>)` to avoid adding a new variant for each vendor-specific column option.
* Support create or replace table
* Support create or replace view
* Simplify create or replace table parser
* Add tests for create or replace external table and materialized view
* Formatting
* Address review comments
* Create error if we didn't see a (external) table or (materialized) view afer create or replace
This implements `DROP [ COLUMN ] [ IF EXISTS ] column_name [ CASCADE ]`
sub-command of `ALTER TABLE`, which is what PostgreSQL supports https://www.postgresql.org/docs/12/sql-altertable.html
(except for the RESTRICT option)
Co-authored-by: Nickolay Ponomarev <asqueella@gmail.com>
Fixes#168 by enabling `DATE` and other keywords to be used as
identifiers when not followed by a string literal.
A "typed string" is our term for generalized version of `DATE '...'`/`TIME '...'`/
`TIMESTAMP '...'` literals, represented as `TypedString { data_type, value }`
in the AST.
Unlike DATE/TIME/TIMESTAMP literals, this is a non-standard extension
supported by PostgreSQL at least.
This is a port of MaterializeInc/materialize#3146
Co-authored-by: Nikhil Benesch <nikhil.benesch@gmail.com>
Co-authored-by: Nickolay Ponomarev <asqueella@gmail.com>
This simplifies codes slightly, removing the need deal with the EOF case explicitly.
The clone kludge in `_ => self.expected("date/time field",
Token::Word(w.clone())))` will become unnecessary once we stop using
a separate match for the keywords, as suggested in
https://github.com/andygrove/sqlparser-rs/pull/193#issuecomment-641607194
Alter INTERVAL to support postgres syntax
This patch updates our INTERVAL implementation such that the Postgres
and Redshfit variation of the syntax is supported: namely that 'leading
field' is optional.
Fixes#177.
This patch provides an initial implemenation of LISTAGG[1]. Notably this
implemenation deviates from ANSI SQL by allowing both WITHIN GROUP and
the delimiter to be optional. We do so because Redshift SQL works this
way and this approach is ultimately more flexible.
Fixes#169.
[1] https://modern-sql.com/feature/listagg
Specifically, `FOREIGN KEY REFERENCES <foreign_table> (<referred_columns>)`
can now be followed by `ON DELETE <referential_action>` and/or by
`ON UPDATE <referential_action>`.
The Ident type was previously an alias for a String. Turn it into a full
fledged struct, so that the parser can preserve the distinction between
identifier value and quote style already made by the tokenizer's Word
structure.
