#SQL quality and diagnostics

JustyBase combines syntax parsing, semantic/schema validation, quality rules, and optional database validation. The diagnostic prefix tells you which layer raised the issue:

Prefix	Meaning	Typical action
`LEX*`	Lexer could not classify the text	Fix an invalid character, literal, or token boundary.
`PAR*`	Parser could not build the expected statement structure	Read the first error and repair the clause around it.
`SQL*`	Parser/semantic validation, often relation, column, type, or identifier scope	Refresh metadata or correct the object/reference.
`NZ*`	SQL quality rule for warehouse safety, style, or performance	Review the warning in the context of workload and policy.
`NZP*`	NZPLSQL procedure rule	Fix block, variable, return, or procedure contract issues.

#Rule families

Safety: destructive statements without a predicate, risky TRUNCATE, and operations that need confirmation.
Correctness: unknown relations/columns, ambiguous references, invalid object qualification, malformed aliases, and type-aware comparisons.
Performance: leading wildcard LIKE, functions in predicates, implicit casts in joins, unnecessary OR, SELECT *, unbounded ordering, and distribution choices.
Style: keyword case, aliases, quoted identifiers, and UNION versus UNION ALL choices.
Types: SQL025/SQL026 warnings compare column metadata with literal classification. They require typed metadata and are not emitted as certain errors without it.
NZPLSQL: procedure structure, language/returns clauses, unmatched blocks, SELECT … INTO, unused variables, OUT assignment, ELSIF, THEN, CASE, and RETURN checks.

#Severity and policy

Set a rule to error, warning, information, hint, or off under justybase.linter.rules. justybase.linter.enabled controls the linter surface; advanced parser-based validation remains the core syntax path. Severity changes the editor and Problems panel, not the SQL sent to the database.

#Before and after

Applying a quick fix for a SQL diagnostic in JustyBase — Quick fixes move from warning to corrected SQL in one step.

#Destructive update

-- Before: NZ003 / safe-execute confirmation
UPDATE SALES.ORDERS SET status = 'CLOSED';

-- After: explicit scope
UPDATE SALES.ORDERS
SET status = 'CLOSED'
WHERE order_date < CURRENT_DATE - 90;

#Wildcard search

-- Before: NZ005 can prevent an index-friendly access path
WHERE customer_name LIKE '%smith';

-- After: choose the intent explicitly
WHERE customer_name ILIKE 'smith%';

#Join type mismatch

-- Before: implicit cast risk
ON fact.customer_id = dim.customer_id_text

-- After: make the conversion and data-quality decision visible
ON CAST(fact.customer_id AS VARCHAR(32)) = dim.customer_id_text

The correct rewrite is workload- and dialect-dependent. A quick fix is a proposal, not permission to execute.

#Quick-fix classes

Class	Behavior
Automatic	The action has an exact, local edit with no database side effect. Review the diff before saving.
Preview required	The action changes a statement or needs DDL/import context; JustyBase shows the proposed SQL or mapping first.
Suggestion only	The rule cannot safely infer intent; use the diagnostic as a review prompt.

Parser-based rename and navigation use token ranges. Regex heuristics are limited to procedure rules that run only when the CST parse cannot provide a reliable result. See the quick-fix coverage matrix for the implementation-level inventory.

#NZPLSQL boundary

The CST procedure path handles structural rules first. A string-body procedure is reparsed with offset mapping. Migrated NZP rules run through the regex fallback only when parsing fails, so a malformed procedure does not silently produce confident semantic results. Always fix the first PAR*/NZP* structural error before interpreting later warnings.

#Database-dependent diagnostics

SQL003/SQL004/SQL007-style relation and qualification checks can be useful without a connection, but column existence and SQL025/SQL026 type comparisons require metadata. The extension-host validator maps catalog FORMAT_TYPE into ColumnInfo.dataType; the LSP provider maps the metadata bridge type into the same contract. If those paths disagree, report the database kind and a minimal query.

#Copilot handoff

The safest AI loop is diagnostic → schema/DDL context → explanation → suggested fix → diff review → user execution. AI SQL Assistant documents what is sent, what is read-only, and where a user confirmation is required.