| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
| |
Implemented a column names specification for a WITH CTE. It requires a
column order derivation. It was not working on `YqlSelect` until PR
https://nda.ya.ru/t/qOC3TNqn7gAcy6 was merged. Also there was a bug
with `UNION` column order inference, which is essential for RECURSIVE
queries. There was compatibility problems with `PgSelect` `UNION`
positional column order semantics, so it was tweaked in `YqlSelect`
mode.
commit_hash:c96192270ca6d51eaa0cb39cf59b65c3951ce3fe
|
| |
|
|
| |
commit_hash:c978cba70a9b24aa2e202caa54dc44f419e19468
|
| |
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes the query <https://nda.ya.ru/t/pYH4zpRH7fw6U4>. If you remove `PRAGMA YqlSelect`,
you'll see that it works on a legacy translation. It is also required to fix PR
<https://nda.ya.ru/t/IZFPoRjS7fw6j8>, because saving a column order makes the bug
visible.
The `nokey` flag of `YqlAgg` callable was renamed to `as_optional` and also an expansion was
fixed by rewriting `MatchType` on `Optional` to `AsOptionalType` to handle `Null` and `pg`
types properly.
commit_hash:dc435437e87006c79bb93984d65931aaf5901af4
|
| |
|
|
|
|
|
|
|
|
| |
Implemented a RECURSIVE WITH CTE translation for YqlSelect. It relies on YqlSelf callable.
As this callable binds to enclosing YqlSelect, the diagnostic that subquery can't reference
an enclosing RECURSIVE WITH CTE was added, it is better, that an existing for pgSQL, where
just "table not found" is emitted. The core idea under RECURSIVE handling is that a
"not ready" CTE entry is added to a namespace (with a `nullptr` node), so it can be recognized.
Checked some interesting test cases and wrote TODOs.
commit_hash:65b8646e0615ae070b04822ce7af24245c203bbe
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The first iteration on WITH CTE support. Only simple CTEs are supported.
Properties:
1. Redefinition is forbidden.
2. Shadowing is supported.
3. Nested WITH CTEs are supported.
Following features are not yet implemented:
1. Column names specification. There is an assertion failure at core because
of an absent column order when translating it like for the `PgSelect`.
2. `RECURSIVE` is postponed. There a more sophisticated translation with
`YqlSelf` callable is required.
Also there was a potential defect because of an empty `Id` result not expected.
commit_hash:759676d8a9d369f7eedd9b87f79455b2f8a9fcff
|
| |
|
|
|
|
|
|
|
|
|
| |
#### `YQL-21321`: Adjust type checking for aggregate traits ✎
- Replaced persistable type check with computable type check during aggregate trait validation to allow more flexible type handling
- Added new test cases to verify the behavior of unused aggregate types and ensure proper error reporting
- Updated canonical data and test artifacts to reflect the changes in type checking logic and error messages
<a href="https://nda.ya.ru/t/qa0kX64r7DqvtN"><font size="2">Autodescription by Yandex Code Assistant</font></a>
commit_hash:e14b06e4a135d9aef8ae78a75c3fb23d983365a6
|
| |
|
|
| |
commit_hash:867eb4209058a877aa60f86931daddf1e25ac600
|
| |
|
|
| |
commit_hash:10c8d54083c7064e63989ad9c341711fe84409b4
|
| |
|
|
|
|
|
| |
- Fixed histogram, (bottom/top)(by)?, topfreq, random_(value/sample) over Null
- Disabled DISTINCT with agg phases
- Refactored code by phase predicates extraction
commit_hash:c82f790224e84e542572d5cd270766e3244b4178
|
| |
|
|
| |
commit_hash:1a546dc230518eb6846155c77db762e4d83281d5
|
| |
|
|
| |
commit_hash:048f5c5f1cf2f507a91c2f4cdb24aec0cc71a004
|
| |
|
|
| |
commit_hash:e420d2d62e3a77af79abbec946f1008c1fa359c6
|
| |
|
|
| |
commit_hash:958a606c2956a1155a487070452acc770455bf66
|
| |
|
|
| |
commit_hash:65d1e3053e157375f7a704f1e635467ded5a7ef2
|
| |
|
|
| |
commit_hash:2cb369cb2091489e28790782fe9e7679d367fbb3
|
| |
|
|
| |
commit_hash:5e471728023b5404426daed232c4b00dbba8026c
|
| |
|
|
| |
commit_hash:617264515b5da826be1327cb91b7fc6f6801a835
|
| |
|
|
| |
commit_hash:934d9bd00af5903f26796cf610789aaa4a7bf39c
|
| |
|
|
| |
commit_hash:72d0f5c9f96875682a71eb11e67783a723112cce
|
| |
|
|
|
|
|
| |
Canonize more tests
Canonize -F "*window*" -F "*tpcds*" tests
Enable window new pipeline by default
commit_hash:6b5900972f1d065c6956e7ee03e734a80e0c6fa5
|
| |
|
|
| |
commit_hash:183e3a952a021c4701c20d6677800f245239d822
|
| |
|
|
| |
commit_hash:13889e43b6156f11945dcee2ab2b59cdb056ce49
|
| |
|
|
| |
commit_hash:cb540d095291ddee3bb60c2d3addee54d23daa1c
|
| |
|
|
|
|
|
|
|
|
| |
Support `FULL JOIN` and fix `RIGHT JOIN` for `YqlSelect`. The
problem with `RIGHT JOIN` is that in a situation, when we have
`[a, b, c, d, e]` on a stack and are going to "perform" `right`,
all of `[a, b, c, d, e]` are marked optional, because the
`inputIndex` is equal to stack size. But we must mark only
`[a, b, c, d]` as optionals.
commit_hash:b78c5bcecba4b736475d3428918c2a3c377373bb
|
| |
|
|
| |
commit_hash:c998ba15db7c4e5137f44904de85755aa7322e8e
|
| |
|
|
| |
commit_hash:af4c180548bf1699ec6afb335167a4220418386c
|
| |
|
|
| |
commit_hash:054395f57e37951159c1184eae2e6d4bc81245f8
|
| |
|
|
|
|
| |
Now `Grouping` function kindly works, but with some known bug,
that will be fixed in the next PR. It gives 4 TPC-DS queries.
commit_hash:e891401453cbd9d9d964f673053b856e55ea8b56
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch adds support for `RowNumber`, `CumeDist`, `NTile`,
`Rank`, `DenseRank`, `PercentRank`, `Lead`, `Lag`, `FirstValue`,
`LastValue` and `NthValue`.
It gives +3 TPC-DS queries and +5 TPC-DS bugs unlocked.
The callable `YqlWin` was introduces, having the following form:
```yqls
(YqlWin
'function_name
'window_name
'() # options
(Void) # type slot
...) # args
```
During the callable type annotation and expansion the same code
for implementation functions substitution is used and shared even
with a pgSQL.
There was an issue with a `listType` for `YqlAgg`, `YqlAggWin`
and `YqlWin`, because it was built incorrectly by referencing
an atom "row" instead of _that_ argument row. It is fixed with
a new "type slot". Now it has 3 states: `(Void)`, which is set
by translator just as a stub, `row` that is set by
`YqlSetItemWrapper` during the `RebuildLambdaColumns`, so then
on the `YqlWin` type annotation it can take this `row` and wrap
it to the `listType`, pass to the function and call for a
repetition, so in a next stage the `resultType` will be ready.
commit_hash:256fc2cb359714a646a308359b340b1715919f5d
|
| |
|
|
| |
commit_hash:a201cbc3a08b7bad471d4d6a15f3dfc3f63d989b
|
| |
|
|
| |
commit_hash:3eed51d012b7cde52449ebd225789e18b8ebd678
|
| |
|
|
| |
commit_hash:f508795113ebbfadfb1bc05529f4af46ae7837d2
|
| |
|
|
| |
commit_hash:198ce0c7875d77fab8082382d915bf1261775f68
|
| |
|
|
|
|
|
|
|
| |
Introduce `YqlAggWin` callable and adapt `PgWindow` as `YqlWindow`.
The `YqlAggWin` is similar to `YqlAgg`, but with window reference (name)
included. Moved `PgWindowWrapper` to `SqlWindowWrapper`. It gives +6 TPC-DS
test cases. Window functions are not yet supported, because they have
separate callables and an extra design is required.
commit_hash:ccef4463fd994779da21514dc9f6f596c511bbfe
|
| |
|
|
| |
commit_hash:820e0aec8856d51711fb71caa225cd7d6d89e7a0
|
| |
|
|
| |
commit_hash:ea278625098c2a9c0ad3bfbab7b981d3cc556020
|
| |
|
|
| |
commit_hash:9bdeca950e4ca97e07b2da163f014c101189c60b
|
| |
|
|
| |
commit_hash:836406699a568f3d53cbd91256d586d37bf8f4a9
|
| |
|
|
| |
commit_hash:9ee085f14c8ce1d9c66c3b1008f71570ea975bf8
|
| |
|
|
| |
commit_hash:9e799ea42a2692eb96ebc792b48e46dd060c182b
|
| |
|
|
|
| |
Add support for automated allocation of secure TmpFolder on YT clusters for each user
commit_hash:2d06fc7616f65d9814c7a5ca33a4da582fb5b016
|
| |
|
|
| |
commit_hash:fd9981b64488aca88d19a45aa956d80eab220cb8
|
| |
|
|
| |
commit_hash:fb14ee5b8cca07451478bfbffcc9a2f0b3d751f5
|
| |
|
|
|
|
| |
`Null` are not yet implemented, because it was hard to fix, will
do it later.
commit_hash:879a892d511a0a2dd7110a728f01f80c260dedbf
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch unlocks a subset of `GROUPING SETS`, `ROLLUP` and `CUBE`
syntax for `YqlSelect`. The `sql` was easy, but I faced problems
at the `core` level. An existing expansion was OK, but type annotation
was changed to correctly annotate `GROUP BY` keys as `Optional`.
```haskell
notNullKeys groupBy =
union [
( intersection [ groupingSet | groupingSet in groupingSets ] )
| groupingSets in groupBy
]
```
It gives 60/100 (+9) TPC-DS queries.
commit_hash:defda3152f10bf16b9e6300294bc23cccfa1329c
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Fixed agg phases for `Top`, `Bottom`, `TopBy` and `BottomBy`. The
`SerializedState` for `Top` and `Bottom` is contains `MaxSize`, so
only the `state` argument is required when it `IsOverState`. `Null`
are not yet implemented, because it was hard to fix, will do it
later. The code is not clean, but to fix it a huge refactoring is
needed, including `aggregate.yqls` and `aggregation.cpp`. Also
YQLs minirun launch target was introduced, because it is easier to
experiment with YQLs directly, rather than tweaking the SQL translator.
commit_hash:47af002f70d9be6db61316e64aa866d4bff34cf4
|
| |
|
|
| |
commit_hash:52386c60ce66e769fe972c72eed82ace0d34242b
|
| |
|
|
| |
commit_hash:1b894cda0775421321d11382600be79c3b5bef3d
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Here was 2 issues:
1. `Init` return status was ignored and so error suppressed.
2. `PreaggregateExpr` was collected twice. On the first attempt it was
replaced with a synthetic column that will be "registered" in a
source later, and on a second attempt it was already that synthetic,
but not yet registered column and so it's init was failing.
commit_hash:4398a075d53d99cbb9b172ebae0a05aff5e2d5a7
|
| |
|
|
| |
commit_hash:63ea284bd7ade74203421fe2a12318f658f058eb
|