ydb/yql/essentials/sql/v1/complete/sql_complete.h, branch CLI_2.21.0

YQL-19747 Complete folder, table and cluster names

2025-05-06T13:04:08Z

--- - Related to `YQL-19747` - On top of https://github.com/ytsaurus/ytsaurus/pull/1253 - Related to https://github.com/ydb-platform/ydb/issues/9056 - Related to https://github.com/vityaman/ydb/issues/14 - Related to https://github.com/vityaman/ydb/issues/35 - Related to https://github.com/vityaman/ydb/issues/40 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1257 commit_hash:0b842abb27184c88b8177beeea29fb1ea86b7a04

YQL-19747: Refactor sql/v1/complete usage of Future and Ptr

2025-04-25T13:30:34Z

- [x] Chained futures, add `CompleteAsync` method (then will migrate the YDB CLI on it). - [x] Removed deadlined and fallback NameService as unused - [x] Annotate thread-safe methods with `const` and use `AtomicSharedPtr` for them. - [x] Move `name` to `name/service` with backward compatibility with the YDB CLI. --- `CompletionEngine` is left thread-unsafe because of the dependency chain `CompletionEngine -> LocalSyntaxAnalysis -> C3Engine` which is thread-unsafe, but readonly indexed data structures such as `Ranking` and `NameService` are annotated with const and distributed via shared pointers. I removed deadlined and fallback name services because the first is stupid the second is ahead of its time and is better to be added later to keep interfaces as minimal as possible. --- The migration on async complete plan: 1. Introduce CompleteAsync 2. Migrate clients on CompleteAsync 3. Make Complete to return Future 4. Migrate clients on Complete 5. Remove CompleteAsync --- - Related to https://github.com/ydb-platform/ydb/issues/9056 - Related to https://github.com/vityaman/ydb/issues/33 - Related to https://github.com/vityaman/ydb/issues/31 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1241 Co-authored-by: vvvv Co-authored-by: vvvv commit_hash:497cc081ab78bebf7354e0acfaa418d936cc8240

YQL-19747 Detect a token at the caret position

2025-04-23T12:27:38Z

When I tried to implement a folder and object names completion at `ID_QUOTED` I faced with a problem, that I actually can't detect, that cursor is at `ID_QUOTED` token because `TCompletionInput::Text` it was cut until the `TCompletionInput::CursorPosition`, therefore at input ``` SELECT * FROM `#` ``` prefix was ``` SELECT * FROM `# ``` and then lexer failed. While we actually want tokenize the whole current statement, `C3` still needs to receive a prefix as input. I tried to tokenize the whole statement and then on input `SELECT Optional<#>` got nothing because `<>` is solid token in the `SQL`. The only way to fix it I found is to cut a query to prefix until the cursor position. BTW, current implementation is not so efficient as we tokenize the input multiple times. Especially `SplitQueryToStatemnts` seems heavy. In future we anyway will parse the whole input so will need to design APIs to receive ready token streams to do statements splitting, for example, just not to do the work twice. ![image](https://github.com/user-attachments/assets/114804d3-f311-4a46-be84-8ed4650bc9dd) So I introduce you the following changes - [x] Select the whole current statement, not just prefix. - [x] Find the token at caret and output no candidates when caret is at `STRING_VALUE`, `DIGIGTS` and so on. - [x] Change `C3` wrapper interface to take `TCompletionInput` to hide an implementation detail that it runs on cut prefix. - [x] `#` annotated queries in unit tests. - [x] Detect `CaretTokenPosition` -- if is it enclosed with a token or between two. - [x] Ensure that `maxErrors` in `ILexer::Tokenize` is positive. Just a tiny bugfix. --- - Related to https://github.com/ytsaurus/ytsaurus/pull/1209 - Related to https://github.com/ydb-platform/ydb/issues/9056 - Related to https://github.com/vityaman/ydb/issues/14 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1225 commit_hash:a434b9888ec8a7356247d63d9f1420e256ae4fca

YQL-19747 Remove default completion engine factory

2025-04-15T11:57:48Z

This is to decouple from the `sql/v1/lexer` implementation. - Related to https://github.com/ydb-platform/ydb/issues/9056 - Related to https://github.com/vityaman/ydb/issues/18 - Following https://github.com/ydb-platform/ydb/pull/16820 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1196 commit_hash:4995f3a7cb14b0b735a48e211111764f65be8033

YQL-19747 Complete select and insert hints

2025-04-09T15:12:05Z

- Related to https://github.com/ydb-platform/ydb/issues/9056 - Related to https://github.com/vityaman/ydb/issues/19 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1189 commit_hash:7f1cb1dcf0617aa2c94c3f2188fc9bd481380252

YQL-19747 Complete after PRAGMA and multi-token names

2025-04-08T13:26:47Z

- [x] Complete after PRAGMA - [x] Complete multi-token names correctly, for example, `yt.` returns only `DisableStrict`, not `yt.DisableStrict` and `DateTime::` returns `Split`, not `DateTime::Split`. I tried to implement it using `CompletedToken` edition, but not all completion environments support candidates with various `contextLen` (`Replxx` does not). So I decided that completions should rewrite only the current token, not sequences. For example, on `DateTime::Spl` rewrite only `Spl`. It makes sense as multi-token names have some namespace separated by a punctuation, so used types only namespace and gets names inside of it. --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1181 commit_hash:9d8967ac43b9348f6dbb53837d92a9dcc9b51f48

YQL-19747 Complete Function Names

2025-03-28T19:36:16Z

- Function names are suggested now - Changed the module structure - Checking ruleIndex independence on mode (ansi | default) via unit tests --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1163 commit_hash:1b1a27d2cff8db663c5c7e8efb57896476823315

YQL-19747 Complete a simple type name

2025-03-28T10:36:24Z

- Related to https://github.com/ydb-platform/ydb/issues/9056 --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1151 commit_hash:6e1e429a2ea805016bf00a1e60b501b7fc8dc8de

YQL-19747 Split statements

2025-03-27T20:42:19Z

When we run completion engine on multi-statement query, where preceding statements are syntactically incorrect, `antlr4-c3` does not return candidates. Running engine only on a current statement provides a best-effort attempt to provide candidates. - Related to https://github.com/ydb-platform/ydb/issues/9056 - Depends on https://github.com/ytsaurus/ytsaurus/pull/1127 (`ELexerFlavor`) --- Pull Request resolved: https://github.com/ytsaurus/ytsaurus/pull/1144 commit_hash:0ced9443a9712191f5420246531f781ca4bc5f42

Intermediate changes

2025-02-28T17:15:53Z

commit_hash:5fc851d2c72810067fe0d407b66535b17de63129