[
https://issues.apache.org/jira/browse/CALCITE-7566?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18085275#comment-18085275
]
Mihai Budiu commented on CALCITE-7566:
--------------------------------------
What about the unparsing side? Do you plan to address that? Could be a separate
issue.
> Support BigQuery-style bare bracket array literals (e.g. [1, 2, 3]) under
> BigQuery conformance
> ----------------------------------------------------------------------------------------------
>
> Key: CALCITE-7566
> URL: https://issues.apache.org/jira/browse/CALCITE-7566
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.41.0
> Reporter: tnakama
> Priority: Minor
> Labels: pull-request-available
>
>
> BigQuery supports a shorthand syntax for array literals using just square
> brackets,
> without the `ARRAY` keyword:
> ```sql
> SELECT [1, 2, 3] AS arr;
> This is documented as equivalent to ARRAY<T>[...] / ARRAY[...] in BigQuery's
> Standard SQL:
>
> https://cloud.google.com/bigquery/docs/reference/standard-sql/lexical#array_literals
> When parsing such queries with Calcite using fun=bigquery and/or
> conformance=BIG_QUERY, the parser fails because it only recognises
> ARRAY[...] / ARRAY(...) constructors.
> Reproduction
> SqlParser.Config config = SqlParser.config()
> .withConformance(SqlConformanceEnum.BIG_QUERY)
> .withLex(Lex.BIG_QUERY);
> SqlParser.create("SELECT [1, 2, 3]", config).parseQuery();
> // → SqlParseException: Encountered "[" at line 1, column 8
> Whereas BigQuery accepts it natively:
> $ bq query --use_legacy_sql=false 'SELECT [1, 2, 3] AS arr'
> +-----------+
> | arr |
> +-----------+
> | [1, 2, 3] |
> +-----------+
> Suggested implementation outline
> In core/src/main/codegen/templates/Parser.jj, the array constructor rule
> (around line 5252) currently requires the ARRAY keyword. Add a
> conformance-gated alternative in AtomicRowExpression (or wherever expression
> atoms are parsed) that, when getConformance().allowBareBracketArrayLiteral()
> (new predicate on SqlConformance) is true, accepts LBRACKET ... RBRACKET and
> emits the same ARRAY_VALUE_CONSTRUCTOR call.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
