fix(parser): stop dropping columns named like constraint keywords

src/core/schema_parser.py

@@ -14,6 +14,9 @@
 
 logger = get_logger(__name__)
 
+# A line that starts with one of these is a table constraint, not a column.
+CONSTRAINT_LINE_RE = re.compile(r"(PRIMARY\s+KEY|FOREIGN\s+KEY|CONSTRAINT|UNIQUE|CHECK|EXCLUDE)\b")
+
 
 class SchemaParser:
     """Parse CREATE TABLE and ALTER TABLE statements from SQL text.
@@ -193,15 +196,9 @@ def _parse_column_line_robust(self, line: str) -> dict | None:
 
         # Skip pure constraint lines
         upper = line.upper().strip()
-        skip_patterns = (
-            "PRIMARY KEY",
-            "FOREIGN KEY",
-            "CONSTRAINT",
-            "UNIQUE",
-            "CHECK",
-            "EXCLUDE",
-        )
-        if upper.startswith(skip_patterns) and "(" in upper:
+        # Anchored on a word boundary so a column named unique_code or
+        # constraint_name is not mistaken for a constraint declaration.
+        if CONSTRAINT_LINE_RE.match(upper) and "(" in upper:
             return None
 
         # Try to extract column name and type

tests/test_schema_parser.py

@@ -190,6 +190,40 @@ def test_table_level_primary_key(self, parser):
         assert len(pks) == 2
         assert {c["name"] for c in pks} == {"a", "b"}
 
+    def test_column_named_like_constraint_keyword(self, parser):
+        """A column whose name starts with a constraint keyword is a column.
+
+        unique_code and constraint_name used to be dropped because the
+        constraint check compared raw prefixes. The designer then showed a
+        table with fewer columns than the SQL declares.
+        """
+        sql = """
+        CREATE TABLE t (
+            id SERIAL PRIMARY KEY,
+            unique_code VARCHAR(10),
+            constraint_name VARCHAR(50),
+            check_flag BOOLEAN,
+            excluded_at TIMESTAMP
+        );
+        """
+        tables, _ = parser.parse_sql(sql)
+        names = [c["name"] for c in tables[0]["columns"]]
+        assert names == ["id", "unique_code", "constraint_name", "check_flag", "excluded_at"]
+
+    def test_table_level_unique_constraint_still_skipped(self, parser):
+        """The keyword fix must not start parsing real constraint lines."""
+        sql = """
+        CREATE TABLE t (
+            id SERIAL PRIMARY KEY,
+            email VARCHAR(200),
+            UNIQUE (email),
+            CONSTRAINT chk_len CHECK (length(email) > 3)
+        );
+        """
+        tables, _ = parser.parse_sql(sql)
+        names = [c["name"] for c in tables[0]["columns"]]
+        assert names == ["id", "email"]
+
     def test_not_null(self, parser):
         sql = "CREATE TABLE t (name VARCHAR(100) NOT NULL);"
         tables, _ = parser.parse_sql(sql)