sql: disable arrays as PKs #50656
Assignees
Labels
A-sql-execution
Relating to SQL execution.
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
Comments
maddyblue
added
C-bug
This was referenced Jun 25, 2020
craig bot
pushed a commit
that referenced
this issue
Jun 29, 2020
50662: sql: disable arrays in forward indexes r=mjibson a=mjibson There are some open questions about inverted index types being used in forward indexes. Disable those for now. Fixes #50656 See #50659 See #17154 Release note (sql change): disable arrays in non-inverted indexes. 50742: sql: granular detection of non-immutable constant casts r=RaduBerinde a=RaduBerinde #### opt: improve tests for parsing timestamps Add more `sem/tree` unit tests for parsing time-related types. In particular: - relative timestamps ('now', 'tomorrow') - TZ variants - timestamps that include timezones for DDate, DTime - cases where the result depends on the current time and/or timezone. In a future change, the parsing functions will also return a flag indicating if the result depends on the context or not; these tests will be updated to check that flag as well. Release note: None #### pgdate: add WithoutTimezone variants The way we parse DTimestamp (and other non-TZ variants) is convoluted: we initially parse it just like a DTimestampTZ, and then we strip away the timezone (adjusting it so the timestamp looks the same as before, except with UTC location). This means that we incorporate the local timezone, even though it is inconsequential. This change adds WithoutTimezone API variants to `pgdate` which avoids using the local timezone. This will allow extending the parsing APIs to indicate whether the result depends on the current time or timezone. Release note: None #### sql: use dummy ParseTimeContext during type-checking We are not supposed to use any context-dependent parsing results during type-checking. As such, use a dummy ParseTimeContext instead of having SemaContext implement the interface. Release note: None #### sql: granular detection of non-immutable constant casts Interpreting a timestamp (and similar types) can depend on the current timezone or even the current time. A recent change prevented these casts from happening during type-checking. Unfortunately, this includes a lot of common cases which aren't actually context-dependent. The most important are dates and timestamps without time zone (except rare cases like 'now' or 'tomorrow'). In these cases, the type-checked expressions won't seem constant and won't be able to be used for partitioning, index predicates, etc. This change improves the parsing functions to return a `dependsOnContext` boolean. When this flag is false, the result is immutable and is replaced during type-checking. The volatility of the version of date_trunc that returns a TIMESTAMPTZ is corrected to Stable. Release note: None Co-authored-by: Matt Jibson <[email protected]> Co-authored-by: Radu Berinde <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The recent addition of an array index encoding had a side effect of enabling them in PKs. This is the first column type that has both inverted and forward index support. However neither the exec nor opt teams have had opportunity to think through the implications of this, and they need to be disabled until all teams sign off.
The text was updated successfully, but these errors were encountered: