wnob commented on code in PR #2973:
URL: https://github.com/apache/calcite/pull/2973#discussion_r1026992374
##########
core/src/main/java/org/apache/calcite/runtime/SqlFunctions.java:
##########
@@ -149,6 +159,40 @@ public class SqlFunctions {
private static final Pattern PATTERN_0_STAR_E = Pattern.compile("0*E");
+ // Date formatter for BigQuery's timestamp literals:
+ //
https://cloud.google.com/bigquery/docs/reference/standard-sql/lexical#timestamp_literals
+ private static final DateTimeFormatter BIG_QUERY_TIMESTAMP_LITERAL_FORMATTER
=
+ new DateTimeFormatterBuilder()
+ // Unlike ISO 8601, BQ only supports years between 1 - 9999,
+ // but can support single-digit month and day parts.
+ .appendValue(ChronoField.YEAR, 4)
+ .appendLiteral('-')
+ .appendValue(ChronoField.MONTH_OF_YEAR, 1, 2, SignStyle.NOT_NEGATIVE)
+ .appendLiteral('-')
+ .appendValue(ChronoField.DAY_OF_MONTH, 1, 2, SignStyle.NOT_NEGATIVE)
+ // Everything after the date is optional. Optional sections can be
nested.
+ .optionalStart()
+ // BQ accepts either a literal 'T' or a space to separate the date
from the time,
+ // so make the 'T' optional but pad with 1 space if it's omitted.
+ .padNext(1, ' ')
+ .optionalStart()
+ .appendLiteral('T')
+ .optionalEnd()
+ // Unlike ISO 8601, BQ can support single-digit hour, minute, and
second parts.
+ .appendValue(ChronoField.HOUR_OF_DAY, 1, 2, SignStyle.NOT_NEGATIVE)
+ .appendLiteral(':')
+ .appendValue(ChronoField.MINUTE_OF_HOUR, 1, 2,
SignStyle.NOT_NEGATIVE)
+ .appendLiteral(':')
+ .appendValue(ChronoField.SECOND_OF_MINUTE, 1, 2,
SignStyle.NOT_NEGATIVE)
+ // ISO 8601 supports up to nanosecond precision, but BQ only up to
microsecond.
+ .optionalStart()
+ .appendFraction(ChronoField.MICRO_OF_SECOND, 0, 6, true)
+ .optionalEnd()
+ .optionalStart()
+ .parseLenient()
+ .appendOffsetId()
+ .toFormatter(Locale.ROOT);
+
Review Comment:
Yes, it would appear that JDK8 had a different implementation for
`DateTimeFormatter` that cannot handle offsets (or at least not in the same way
as later versions). I'm inclined to switch to regex-based conversion, which
will require more logic but be more widely compatible. Thoughts?
##########
site/_docs/reference.md:
##########
@@ -2596,7 +2596,16 @@ semantics.
| p | CONVERT_TIMEZONE(tz1, tz2, datetime) | Converts the timezone
of *datetime* from *tz1* to *tz2*
| b | CURRENT_DATETIME([timezone]) | Returns the current
time as a TIMESTAMP from *timezone*
| m | DAYNAME(datetime) | Returns the name, in
the connection's locale, of the weekday in *datetime*; for example, it returns
'星期日' for both DATE '2020-02-10' and TIMESTAMP '2020-02-10 10:10:10'
-| b | DATE(string) | Equivalent to
`CAST(string AS DATE)`
+| b | DATE(integer, integer, integer) | Returns a date object
given year, month, and day
+| b | DATE(timestamp) | Extracts the UTC date
from a timestamp
+| b | DATE(timestamp, string) | Extracts the date from
a timestamp, in a given time zone
+| b | TIMESTAMP(string) | Equivalent to
`CAST(string AS TIMESTAMP)`
+| b | TIMESTAMP(string, string) | Equivalent to
`CAST(string AS TIMESTAMP)`, converted to a given time zone
+| b | TIMESTAMP(date) | Convert a UTC date to a
timestamp (at midnight)
+| b | TIMESTAMP(date, string) | Convert a date to a
timestamp (at midnight), in a given time zone
+| b | TIME(integer, integer, integer) | Returns a time object
given hour, minute, and second
+| b | TIME(timestamp) | Extracts the UTC time
from a timestamp
Review Comment:
Reworded for clarity.
##########
babel/src/main/codegen/includes/parserImpls.ftl:
##########
@@ -42,6 +42,46 @@ SqlNode DateFunctionCall() :
}
}
+SqlNode TimestampFunctionCall() :
+{
+ final SqlFunctionCategory funcType =
SqlFunctionCategory.USER_DEFINED_FUNCTION;
Review Comment:
Ya this was just copied from the existing `DateFunctionCall`. I assumed this
vaguely meant library function since there is no obvious "library function"
value in the `SqlFunctionCategory` enum. Perhaps these should all be in the
`TIMEDATE` category. I suppose since this is the babel parser, there's no real
concept of a library function.
BTW, is my assessment of babel in my above comment with Oliver correct?
Should we be putting these productions in the core parser instead (or in
addition to here)?
##########
babel/src/test/resources/sql/big-query.iq:
##########
@@ -1051,29 +1051,98 @@ select unix_date(timestamp '2008-12-25') as d;
!ok
-# DATE
-# 'date(x) is shorthand for 'cast(x as date)'
-select date('1970-01-01') as d;
+# DATE(year, month, day)
+select date(2022, 11, 15) as d;
+------------+
| d |
+------------+
-| 1970-01-01 |
+| 2022-11-15 |
+------------+
(1 row)
!ok
-!if (false) {
-select date(cast(null as varchar(10))) as d;
-+---+
-| D |
-+---+
-| |
-+---+
+# TIMESTAMP(string)
Review Comment:
Cool, I think the most recent commit has what you asked for. The BigQuery
docs aren't completely self-consistent style-wise but I copied verbatim. Also,
the 3 latter tests involve a combination of different functions, so just kept
the original descriptive comments there.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
