sql: Make INT an alias for INT8 #32831
Conversation
|
This change is |
pkg/sql/sem/tree/col_types_test.go
Outdated
| {"INTEGER", "CREATE TABLE a (b INT8)", &coltypes.TInt{Width: 64}}, | ||
| } | ||
| for i, d := range testData { | ||
| sql := fmt.Sprintf("CREATE TABLE a (b %s)", d.str) |
There was a problem hiding this comment.
I encourage subtests here. That will allow using t.Fatal instead of t.Error and removing the various continues.
There was a problem hiding this comment.
Very nicely done. Just a few minor comments remaining.
Reviewed 130 of 130 files at r1.
Reviewable status:complete! 0 of 0 LGTMs obtained
pkg/sql/serial.go, line 74 at r1 (raw file):
// If unique_rowid() or virtual sequences are requested, we have // no choice but to use the full-width integer type, no matter // which serial size was requested, otherwise the values will not fit.
You need to add a new TODO here; when the type SERIAL is specified and the default int size is 4, we may want to flip the behavior here and force the use of a real SQL sequence.
Please also file a follow-up issue so we can investigate this further with example client apps interested in this change.
pkg/sql/distsqlplan/physical_plan.go, line 906 at r1 (raw file):
func equivalentTypes(c, other *sqlbase.ColumnType) bool { // Convert a "naked" INT type to INT8 if c.SemanticType == sqlbase.ColumnType_INT && c.Width == 0 {
lhs := *c
then modify lhs in-place
(this avoids escaping to the heap. Please check the result with the escape analysis tool. In case of doubt double-check with the original author of the code.)
pkg/sql/sem/tree/normalize_test.go, line 44 at r1 (raw file):
{`(a)`, `a`}, {`((((a))))`, `a`}, {`CAST(NULL AS INT2)`, `CAST(NULL AS INT8)`},
Add a comment to explain why this happens.
pkg/sql/sem/tree/type_check_test.go, line 45 at r1 (raw file):
{`NULL || 'hello'`, `NULL`}, {`NULL || 'hello'::bytes`, `NULL`}, {`NULL::int4`, `NULL::INT4`},
this one is ok
There was a problem hiding this comment.
@BramGruneir, whom I forgot to invite to the review.
Reviewable status:
pkg/sql/serial.go, line 74 at r1 (raw file):
Previously, knz (kena) wrote…
You need to add a new TODO here; when the type SERIAL is specified and the default int size is 4, we may want to flip the behavior here and force the use of a real SQL sequence.
Please also file a follow-up issue so we can investigate this further with example client apps interested in this change.
Done.
pkg/sql/distsqlplan/physical_plan.go, line 906 at r1 (raw file):
Previously, knz (kena) wrote…
lhs := *cthen modify
lhsin-place(this avoids escaping to the heap. Please check the result with the escape analysis tool. In case of doubt double-check with the original author of the code.)
Done.
pkg/sql/sem/tree/col_types_test.go, line 106 at r1 (raw file):
Previously, mjibson (Matt Jibson) wrote…
I encourage subtests here. That will allow using t.Fatal instead of t.Error and removing the various continues.
Done.
pkg/sql/sem/tree/normalize_test.go, line 44 at r1 (raw file):
Previously, knz (kena) wrote…
Add a comment to explain why this happens.
Done.
There was a problem hiding this comment.
Reviewed 24 of 24 files at r2.
Reviewable status:
pkg/sql/serial.go, line 76 at r2 (raw file):
// which serial size was requested, otherwise the values will not fit. // // TODO: Follow up with https://github.com/cockroachdb/cockroach/issues/32534
We have a linter it should be TODO(bob): ...
This is the first in a series of changes to allow INT to eventually be redefined as INT4. Currently, a "naked" INT is generally treated as a 64-bit int type, but INT is separate from INT8. This change eliminates INT as a distinct type in favor of always using INT8. This is analogous to how we treat FLOAT/FLOAT8. The coltypes.Int global has been removed and we should no longer produce type objects that have a zero Width. The existing interpretation of zero-Width types as 64-bit values will continue indefinitely to preserve backwards-compatibility with existing descriptors. In general, the changes to tests are due to the widespread use of INT as an arbitary datatype. Except in a few cases where we are testing how "INT" is interpreted as a datatype, the tests have been changed to simply use INT8, to avoid another major round of text fixups as work on #26925 continues. Release note (sql change): The INT type is now treated as an alias for INT8.
|
bors r+ |
32831: sql: Make INT an alias for INT8 r=bobvawter a=bobvawter This is the first in a series of changes to allow INT to eventually be redefined as INT4. Currently, a "naked" INT is generally treated as a 64-bit int type, but INT is separate from INT8. This change eliminates INT as a distinct type in favor of always using INT8. This is analogous to how we treat FLOAT/FLOAT8. The coltypes.Int global has been removed and we should no longer produce type objects that have a zero Width. The existing interpretation of zero-Width types as 64-bit values will continue indefinitely to preserve backwards-compatibility with existing descriptors. In general, the changes to tests are due to the widespread use of INT as an arbitary datatype. Except in a few cases where we are testing how "INT" is interpreted as a datatype, the tests have been changed to simply use INT8, to avoid another major round of text fixups as work on #26925 continues. Release note (sql change): The INT type is now treated as an alias for INT8. Co-authored-by: Bob Vawter <[email protected]>
Build succeeded |
32848: sql: Add default_int_size to control INT alias r=bobvawter a=bobvawter This is a follow-up from #32831 to support #26925. The combination of `default_int_size` interacts with `experimental_serial_normalization`. Sequence types that are ultimately based on `unique_rowid()` must always be INT8 types in order to hold the returned values. Only in the case where `default_int_size=4 experimental_serial_normalization='sql_sequence'` will an INT4 column be created in response to a SERIAL type. Release note (sql change): A new session variable `default_int_size` and cluster setting `sql.defaults.default_int_size` have been added to control how the INT and SERIAL types are interpreted. The default value, 8, causes the INT and SERIAL types to be interpreted as aliases for INT8 and SERIAL8, which have been the historical defaults for CockroachDB. PostgreSQL clients that expect INT and SERIAL to be 32-bit values, can set `default_int_size` to 4, which will cause INT and SERIAL to be aliases for INT4 and SERIAL4. Please note that due to issue #32846, `SET default_int_size` does not take effect until the next statement batch is executed. Co-authored-by: Bob Vawter <[email protected]>
|
Don't conflate SERIAL and INT. Reserve the freedom to implement SERIAL as a (locally!) ever-increasing (globally!) unique value (with gaps), without restricting it to INT, or to UUID. |
|
@druud this behavior is not changed. As-is SERIAL will continue to map to |
This is the first in a series of changes to allow INT to eventually be
redefined as INT4.
Currently, a "naked" INT is generally treated as a 64-bit int type, but INT is
separate from INT8. This change eliminates INT as a distinct type in favor of
always using INT8. This is analogous to how we treat FLOAT/FLOAT8.
The coltypes.Int global has been removed and we should no longer produce type
objects that have a zero Width. The existing interpretation of zero-Width types
as 64-bit values will continue indefinitely to preserve backwards-compatibility
with existing descriptors.
In general, the changes to tests are due to the widespread use of INT as an
arbitary datatype. Except in a few cases where we are testing how "INT" is
interpreted as a datatype, the tests have been changed to simply use INT8, to
avoid another major round of text fixups as work on #26925 continues.
Release note (sql change): The INT type is now treated as an alias for INT8.