PostgreSQL: COMMENT ON CONSTRAINT / OPERATOR / RULE#2324
Closed
fmguerreiro wants to merge 78 commits into
Closed
Conversation
…R TABLE operations (#1) PostgreSQL supports FORCE ROW LEVEL SECURITY and NO FORCE ROW LEVEL SECURITY as ALTER TABLE operations. Add parsing support for both variants.
feat: parse EXCLUDE constraints for PostgreSQL
…m-main # Conflicts: # Cargo.toml # src/ast/ddl.rs # src/ast/helpers/stmt_create_table.rs # src/ast/mod.rs # src/ast/spans.rs # src/ast/table_constraints.rs # src/parser/mod.rs # tests/sqlparser_postgres.rs
chore: merge upstream apache/datafusion-sqlparser-rs main
- Add HANDLER and VALIDATOR keywords to the keyword list - Add FdwRoutineClause enum for HANDLER/NO HANDLER and VALIDATOR/NO VALIDATOR clauses - Add CreateForeignDataWrapper struct for CREATE FOREIGN DATA WRAPPER - Add CreateForeignTable struct for CREATE FOREIGN TABLE - Export new types from ast::mod and add Statement variants - Add spans.rs coverage returning Span::empty() for the new variants
- Dispatch on FOREIGN keyword in parse_create, branching on DATA WRAPPER or TABLE - parse_create_foreign_data_wrapper: parses optional HANDLER/NO HANDLER, VALIDATOR/NO VALIDATOR, and OPTIONS clauses - parse_create_foreign_table: parses IF NOT EXISTS, column list via parse_columns, required SERVER name, and optional OPTIONS clause
Round-trip tests via pg().verified_stmt for: - FDW: name-only, HANDLER, NO HANDLER, NO VALIDATOR, combined HANDLER+VALIDATOR+OPTIONS - FOREIGN TABLE: basic columns+SERVER, IF NOT EXISTS, table-level OPTIONS
Add Statement::CreateAggregate, CreateAggregate struct, CreateAggregateOption enum, and AggregateModifyKind enum to represent PostgreSQL CREATE AGGREGATE DDL. Options are stored as a typed enum covering all documented parameters (SFUNC, STYPE, FINALFUNC, PARALLEL, moving-aggregate variants, etc.).
Wire AGGREGATE into the CREATE dispatch (before the or_replace error branch so CREATE OR REPLACE AGGREGATE is accepted). parse_create_aggregate parses the name, argument-type list, and the options block. Each recognised option keyword dispatches to parse_create_aggregate_option which produces the typed CreateAggregateOption variant.
Three tests covering: basic old-style aggregate (SFUNC/STYPE/FINALFUNC/INITCOND), CREATE OR REPLACE with PARALLEL = SAFE, and moving-aggregate options (MSFUNC/MINVFUNC/MSTYPE/MFINALFUNC_EXTRA/MFINALFUNC_MODIFY). All use pg().verified_stmt() to assert parse-then-display round-trips identically.
feat: parse CREATE TEXT SEARCH CONFIGURATION/DICTIONARY/PARSER/TEMPLATE
# Conflicts: # src/ast/ddl.rs # src/ast/mod.rs
# Conflicts: # src/ast/ddl.rs # src/ast/mod.rs # src/parser/mod.rs
feat: parse CREATE AGGREGATE
# Conflicts: # src/ast/ddl.rs # src/ast/mod.rs # tests/sqlparser_postgres.rs
feat: parse CREATE STATISTICS, CREATE ACCESS METHOD, CREATE EVENT TRIGGER, CREATE TRANSFORM
# Conflicts: # src/ast/ddl.rs # src/ast/mod.rs # src/ast/spans.rs # src/parser/mod.rs # tests/sqlparser_postgres.rs
feat: parse CREATE CAST, CREATE CONVERSION, CREATE LANGUAGE, CREATE RULE
chore: bump version to 0.60.10
PRs #14, #15, and #16 each added new AST types that collide in the same places in src/ast/ddl.rs, src/ast/mod.rs, src/parser/mod.rs, and tests/sqlparser_postgres.rs. Several mid-match-arm and mid-function-body concat resolutions during the three rebase-and-merge passes left fn / impl / match blocks missing their closing braces and left the ddl re-export list with duplicate entries. This restores compile: - Adds missing closing braces for CastFunctionKind, StatisticsKind, CreateRule / CreateStatistics / CreateEventTrigger / CreateTransform / CreateLanguage Display impls and parse fns, plus the CreateTransform From-to-Statement impl. - Dedupes and alphabetizes the `pub use self::ddl::{ ... }` list in src/ast/mod.rs. - Adds missing doc comments on CastFunctionKind fields and RuleEvent variants that #[warn(missing_docs)] requires. Unblocks 0.60.10 publish.
fix: close delimiter gaps left by concat merges
PostgreSQL's pg_dump emits WITH NO DATA on unpopulated materialized views. Add a with_data: Option<bool> field to CreateView and parse the clause after the query body. Bump to 0.60.11.
This was an accidentally-committed gitlink (160000 mode) with no corresponding entry in .gitmodules. It does not break git itself but causes any cargo git-dep consumer to fail cloning the repo: failed to update submodule `.worktrees/fix-pr-2307-ci` no URL configured for submodule '.worktrees/fix-pr-2307-ci' Same class of residue that 837b5a0 on feat/exclude-constraint-upstream cleaned up, but that commit never reached main so the gitlink still ships on every descendant branch.
feat(parser): parse WITH [NO] DATA on CREATE MATERIALIZED VIEW
Add AlterTableOperation::AttachPartitionOf and DetachPartitionOf variants
for PostgreSQL declarative partition DDL:
ALTER TABLE parent ATTACH PARTITION child { FOR VALUES <spec> | DEFAULT }
ALTER TABLE parent DETACH PARTITION child [ CONCURRENTLY | FINALIZE ]
These are distinct from the existing ClickHouse AttachPartition/DetachPartition
variants which use the PART|PARTITION <expr> syntax.
- Add FINALIZE keyword to the keyword list
- Add span impls for the new variants
- Add parser tests covering range, list, hash, default attach and
plain, concurrently, finalize detach
- Bump version to 0.60.12
…h-partition feat: parse ALTER TABLE ATTACH/DETACH PARTITION for PostgreSQL
FULLTEXT and SPATIAL are MySQL-specific table constraint keywords. When parse_optional_table_constraint encountered them in a PostgreSQL context it would consume the token, fall through to the _ arm, and call prev_token() to restore it — but the consume-then-backtrack round-trip left room for a subtle parser-state bug that caused the next call to parse_data_type to see 'fulltext' instead of the actual column type. Fix: add an explicit peek-based guard at the top of parse_optional_table_constraint so that for dialects other than MySQL / Generic the FULLTEXT / SPATIAL tokens are never consumed at all. The caller (parse_columns) then correctly falls through to parse_column_def and treats the word as an ordinary column identifier. Also add a parser unit test covering a tsvector column literally named 'fulltext' (as in the pagila DVD-rental schema) and a GiST index that references the same column by name. Bump version to 0.60.13.
fix(parser): guard MySQL FULLTEXT/SPATIAL table constraint on dialect
…TE FUNCTION
TriggerExecBody previously reused FunctionDesc (and parse_function_desc),
which parsed arguments as CREATE-FUNCTION-style parameter declarations
(argname argtype DEFAULT expr). Any trigger that passes literal arguments
such as tsvector_update_trigger('fulltext', 'pg_catalog.english', ...) would
fail with "Expected: a data type name, found: 'fulltext'".
Replace the func_desc: FunctionDesc field with func_name: ObjectName and
args: Option<Vec<Expr>>, and replace the parse_function_desc() call in
parse_trigger_exec_body with parse_object_name() + parse_comma_separated
(Parser::parse_expr). This treats args as call-site expressions, matching
PostgreSQL semantics. The None/Some(vec![]) distinction preserves whether
parentheses were written at all.
Bump version to 0.60.14.
fix(parser): accept call-site expression args in CREATE TRIGGER EXECUTE FUNCTION
…est) (#23) * chore(tests): restore green main; fix clippy and keyword sort regressions `cargo test --all-features` and `cargo clippy --all-targets --all-features -- -D warnings` were both failing on `main`; CI never ran on the fork to catch it. This commit makes them green so subsequent PRs can be verified. Test compile errors: - `Statement::AlterTable` was refactored from struct to tuple form carrying `AlterTable`. Update 10 ATTACH/DETACH PARTITION test sites in `sqlparser_postgres.rs` to match the new shape. - `CreateView` gained a `with_data` field (added with the `WITH [NO] DATA` parser support). Add `with_data: _,` to 6 destructure sites in `sqlparser_common.rs`. Test runtime failures: - `all_keywords_sorted`: `CONFIGURATION`, `INLINE`, and `TRANSFORM` were inserted out of ASCII order in `keywords.rs`; move each to its alphabetical slot. - `alter_procedure_rename` / `alter_procedure_set_search_path` / `parse_fulltext_column_and_index_in_postgres`: `verified_stmt` round-trip fails because Display canonicalises type names and access methods to uppercase. Update the input SQL to match. Lib clippy regressions (forbidden via `#![forbid(clippy::unreachable)]` and `-D warnings`): - `parse_create_aggregate_options` PARALLEL arm: replace `_ => unreachable!()` with an explicit `Internal parser error` so `clippy::unreachable` is satisfied. - `parse_create_aggregate_args`: drop the let-and-return. - `parse_create_rule` INSTEAD/ALSO: collapse the always-false else-if branches into a single optional-keyword consume. Doctest: - `Statement::CreateStatistics` doc block was missing its opening ```sql fence. Spans: - Drop the duplicate `ForceRowLevelSecurity` / `NoForceRowLevelSecurity` arms in `AlterTableOperation::span()` (warnings, but `-D warnings` blocks clippy). * chore(ci): fix lint, no-std, RAT, and benchmark-lint regressions CI on `main` was failing five jobs (test/lint/compile-no-std/RAT/ benchmark-lint). The first commit on this branch fixed test compile + runtime + a few clippy and doctest issues found locally. CI surfaced the rest. This commit covers them. `lint` (clippy 1.95 — 8 collapsible_match errors) - Fold inner `if` blocks into match guards in `parse_sql` (END token delimiter), `parse_compound_field` (period-qualified identifier), the DO/loop sequence terminator, and the five `HiveRowFormat::DELIMITED` arms (FIELDS / COLLECTION / MAP / LINES / NULL DEFINED AS). `compile-no-std` (`thumbv6m-none-eabi`) - `src/ast/table_constraints.rs` introduced an EXCLUDE-constraint type with `pub operator: String` but did not import `String` in the `#[cfg(not(feature = "std"))]` alloc block. Add `string::String`. `Release Audit Tool (RAT)` - Add `CLAUDE.md` to `dev/release/rat_exclude_files.txt` next to the existing `AGENTS.md` exclusion. Both are agent-guideline files that do not need an Apache header. `benchmark-lint` - `sqlparser_bench/Cargo.toml` referenced the path-dep as `sqlparser` but the fork's package was renamed to `pgmold-sqlparser`. Add the `package = "pgmold-sqlparser"` rename so the bench resolves. Verified locally on rustc 1.95.0: cargo fmt --all -- --check cargo clippy --all-targets --all-features -- -D warnings cargo test --all-features cargo check --no-default-features --target thumbv6m-none-eabi (cd sqlparser_bench && cargo clippy --all-targets --all-features -- -D warnings)
…AIN (#24) `GRANT USAGE ON TYPE foo TO role` and `GRANT ON DOMAIN d TO role` are valid PostgreSQL grammar but failed to parse — `parse_grant_target` had no `TYPE` or `DOMAIN` keyword, and `GrantObjects` lacked variants to carry them. Add `GrantObjects::Types(Vec<ObjectName>)` and `GrantObjects::Domains(Vec<ObjectName>)`, along with their `Display` arms, the `Keyword::TYPE` / `Keyword::DOMAIN` arms in `parse_grant_deny_revoke_privileges_objects`, and round-trip tests in `parse_grant`. Verified: cargo fmt --all -- --check cargo clippy --all-targets --all-features -- -D warnings cargo test --all-features parse_grant
…-quoted bodies (#25) * PostgreSQL: extend COMMENT ON for trigger/aggregate/policy and dollar-quoted bodies Adds Trigger, Aggregate, and Policy variants to CommentObject and extends Statement::Comment with optional `arguments: Option<Vec<DataType>>` (for FUNCTION/PROCEDURE/AGGREGATE overload signatures) and `relation: Option<ObjectName>` (for the `ON <table>` tail of TRIGGER and POLICY). Parser now accepts: - COMMENT ON AGGREGATE name(arg, ...) IS '...' - COMMENT ON TRIGGER name ON table IS '...' - COMMENT ON POLICY name ON table IS '...' - COMMENT ON FUNCTION/PROCEDURE name(arg, ...) IS '...' parse_literal_string now accepts Token::DollarQuotedString in the PostgreSQL/Generic dialects, enabling COMMENT ON ... IS $$body$$ and $tag$body$tag$. NOTE: CommentObject and Statement::Comment are not #[non_exhaustive], so new variants and new struct fields are a breaking change for downstream consumers that pattern-match exhaustively. Recommend bumping to 0.61.0 for the next release rather than a 0.60.x patch. * PostgreSQL: require argument list for COMMENT ON AGGREGATE, escape quotes in Display - Reject bare `COMMENT ON AGGREGATE foo IS '...'`; PostgreSQL requires an argument list as part of the aggregate's identity (distinguishes ordinary vs ordered-set forms). FUNCTION and PROCEDURE remain tolerant of the bare form for backward compatibility. - Escape embedded single quotes in the Statement::Comment Display impl via value::escape_single_quote_string. This was latent but became reachable with dollar-quoted parsing (e.g. $$it's$$ previously round-tripped to invalid SQL `'it's'`). - Add negative parser test for the AGGREGATE guard and a round-trip test for a dollar-quoted body containing a single quote.
…nt::AlterDefaultPrivileges (#26) * PostgreSQL: AlterTypeOperation owner/schema/attribute ops and Statement::AlterDefaultPrivileges Postgres rejects `ALTER TYPE ... OWNER TO ...`, `ALTER TYPE ... SET SCHEMA ...`, `ALTER TYPE ... { ADD | DROP | ALTER | RENAME } ATTRIBUTE ...`, and every `ALTER DEFAULT PRIVILEGES ...` form because the AST has no shape for them. This forces downstreams (e.g. pgmold) to strip those statements before parsing. `AlterTypeOperation` gains six variants (`OwnerTo`, `SetSchema`, `AddAttribute`, `DropAttribute`, `AlterAttribute`, `RenameAttribute`). The parser branches accept the matching grammar including optional `IF [NOT] EXISTS`, `COLLATE`, and `CASCADE | RESTRICT`. A new `Statement::AlterDefaultPrivileges` carries `for_roles`, `in_schemas`, and an `AlterDefaultPrivilegesAction { Grant | Revoke }` body that reuses `Privileges` and `Grantee`. The `ON <kind>` target uses a new `AlterDefaultPrivilegesObjectType` enum (`TABLES | SEQUENCES | FUNCTIONS | ROUTINES | TYPES | SCHEMAS`). New keywords: `ATTRIBUTE`, `ROUTINES`, `TYPES`. The `parse_alter` dispatch now accepts `DEFAULT` as the leading word. Note: both `AlterTypeOperation` and `Statement` are not `non_exhaustive`, so adding variants is technically a breaking change for downstream consumers that match exhaustively without `_`. Recommend bumping the next release to 0.61.0. Verified: cargo fmt --all cargo clippy --all-targets --all-features -- -D warnings cargo test --all-features cargo check --no-default-features --target thumbv6m-none-eabi (cd sqlparser_bench && cargo clippy --all-targets --all-features -- -D warnings) * PostgreSQL: fix ALTER TYPE ADD ATTRIBUTE grammar, Grantee PUBLIC Display Review fixes for PR #26: 1. Remove IF NOT EXISTS from AlterTypeOperation::AddAttribute. PostgreSQL's grammar for ADD ATTRIBUTE does not include IF NOT EXISTS; only ADD VALUE (for enum types) does. The parser was accepting SQL that PostgreSQL itself rejects. 2. Fix pre-existing trailing-space bug in Grantee Display: GranteesType::Public wrote "PUBLIC " (with trailing space) then fell through to the name-formatting path, yielding "PUBLIC " for bare PUBLIC grantees. Early-return with "PUBLIC" instead. Adds regression tests for GRANT ... TO PUBLIC and ALTER DEFAULT PRIVILEGES ... TO PUBLIC round-trips. 3. Document that AlterDefaultPrivileges::for_roles is keyword-agnostic: the parser accepts both FOR ROLE and FOR USER (PG synonyms) but Display canonicalises to FOR ROLE.
…rgname/VARIADIC (#28) Postgres permits argmode (IN/OUT/INOUT/VARIADIC) and argname tokens ahead of each argument type in COMMENT ON FUNCTION / PROCEDURE / AGGREGATE; they are ignored for identity resolution but appear in real-world dumps and consumer schema files. The previous parser used bare parse_data_type for the argument list, which hard-failed on those shapes. Switch the function-like arms of parse_comment to parse_function_arg and project to data_type, since Statement::Comment.arguments stores a plain Vec<DataType> and the modes/names are intentionally discarded for identity matching. Add tests covering argname, IN/OUT/INOUT, VARIADIC, and AGGREGATE.
…30) * PostgreSQL: code-quality cleanup for COMMENT, AlterType, AlterDefaultPrivileges - CommentObject: add `keyword_str` and `from_keyword` helpers; collapse the 18-arm `parse_comment` match into a single lookup, with `MATERIALIZED VIEW` preserved as the only multi-keyword case. Reuse `keyword_str` for `Display`. - Statement::Comment Display: drop the stray `;` after `IF EXISTS` write. - ALTER TYPE ... ALTER ATTRIBUTE: comment the `let _ = parse_keywords([SET, DATA])` so it reads as canonicalization rather than a swallowed result. - AlterDefaultPrivilegesAction Display: extract a shared `write_privileges_body` helper for the GRANT/REVOKE body shape. - AlterDefaultPrivileges: remove the redundant `Spanned` impl that returned `Span::empty()`; the `Statement::AlterDefaultPrivileges` arm in `spans.rs` already covers it. - AlterTypeOperation attribute Display arms: use `display_option_spaced` for the trailing `drop_behavior`, dropping four near-identical `if let Some` blocks. - tests: drop a duplicate empty-arg `verified_stmt` between `parse_comment_on_aggregate` and `parse_comment_on_function_with_arg_types`. No behavior change. * fix: drop spurious import; display_option_spaced lives in ddl.rs
Aligns the partner-table field name with sibling AST nodes (AlterPolicy.table_name, CreatePolicy.table_name, DropPolicy.table_name, AlterTrigger.table_name). Breaking change on the public AST. Pure mechanical rename; no behavior change. Display impl, parser constructor, and existing COMMENT ON TRIGGER / POLICY / AGGREGATE / FUNCTION / COLLATION tests updated to match.
Adds parser + AST coverage for the three remaining COMMENT ON shapes
that pgmold-sqlparser had been silently rejecting:
- `COMMENT ON CONSTRAINT name ON [DOMAIN] target IS '…'`
- `COMMENT ON OPERATOR name (left, right) IS '…'` (each side may be NONE)
- `COMMENT ON RULE name ON target IS '…'`
`Statement::Comment` gains two fields: `operator_args:
Option<CommentOperatorArgs>` (operand types for OPERATOR, modeled as
`Option<DataType>` per side so unary `NONE` is preserved), and
`on_domain: bool` (set when CONSTRAINT uses the `ON DOMAIN <domain>`
form). The existing `arguments` and `table_name` fields keep their
prior semantics for FUNCTION/PROCEDURE/AGGREGATE and TRIGGER/POLICY.
Refactors `parse_drop_operator_signature` to share the new
`parse_operator_arg_type_or_none` helper with `parse_comment`, since
both parse the `{ DataType | NONE }` slot shape.
Bumps fork to 0.63.0 and adds changelog/0.63.0.md.
Contributor
Author
|
Created against the wrong remote — closing. This is fork-only work targeting fmguerreiro/datafusion-sqlparser-rs (the pgmold-sqlparser publishing fork). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds parser + AST coverage for the three remaining
COMMENT ON …shapes that the fork had been silently rejecting:COMMENT ON CONSTRAINT name ON [DOMAIN] target IS '…'COMMENT ON OPERATOR name (left, right) IS '…'— each side may beNONEfor unary operatorsCOMMENT ON RULE name ON target IS '…'Changes
CommentObjectgainsConstraint,Operator,Rulevariants inkeyword_strandfrom_keyword.CommentOperatorArgs { left: Option<DataType>, right: Option<DataType> }carries the operand signature forOPERATOR, withDisplayrenderingNONEfor unary slots.Statement::Commentgains two fields:operator_args: Option<CommentOperatorArgs>andon_domain: bool(only meaningful forConstraint). Existing destructures will need to add the two fields.parse_commentdispatches the new variants:OPERATORusesparse_operator_name()(so+,myschema.@@, etc. round-trip) and a mandatory(slot, slot)signature.CONSTRAINTandRULErequire anON <relation>tail;CONSTRAINTadditionally acceptsON DOMAIN <domain>.parse_drop_operator_signatureto share a newparse_operator_arg_type_or_none()helper withparse_comment.0.63.0; addschangelog/0.63.0.md.Breaking changes
Statement::Commentaddsoperator_argsandon_domainfields. Consumers that destructure all fields explicitly must add them; those using..rest patterns are unaffected. Documented inchangelog/0.63.0.md.Test plan
cargo test --package pgmold-sqlparser(CI)tests/sqlparser_postgres.rs:parse_comment_on_constraint_on_table(binary form +IS NULL+IF EXISTS)parse_comment_on_constraint_on_domain(unqualified + schema-qualified domain)parse_comment_on_constraint_requires_relation_tail(error case)parse_comment_on_operator_binary(public.+(INTEGER, INTEGER))parse_comment_on_operator_prefix_left_none(unary prefix,NONEleft)parse_comment_on_operator_postfix_right_none(unary postfix,NONEright)parse_comment_on_operator_requires_argument_list(error case)parse_comment_on_rule+IS NULLparse_comment_on_rule_requires_relation_tail(error case)Statement::Commentdestructures intests/sqlparser_postgres.rsto include the two new fields.