Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/backend/catalog/Makefile`:
- Line 60: The install copies oracle_compat_views.sql but initdb never executes
it; update src/bin/initdb/initdb.c to call setup_run_file() for
"oracle_compat_views.sql" just like the existing calls for "system_views.sql"
and "information_schema.sql" so the CREATE OR REPLACE VIEW in
oracle_compat_views.sql is run during database initialization; locate the initdb
function that invokes setup_run_file() for other catalog SQL scripts and add a
matching setup_run_file(...) invocation for oracle_compat_views.sql (same
pattern and error handling as the existing calls).

In `@src/backend/catalog/oracle_compat_views.sql`:
- Around line 39-45: The inner join to information_schema.key_column_usage
(alias kcu) is dropping CHECK constraints; change that JOIN to a LEFT JOIN, add
a second LEFT JOIN to information_schema.constraint_column_usage (alias ccu) to
surface columns for CHECK constraints, and update the column selection (where
the column name is chosen) to use COALESCE(kcu.column_name, ccu.column_name) so
CHECK constraints are preserved while keeping existing PK/FK/UNIQUE behavior
(refer to aliases tc, kcu, ccu and the column selection that currently picks
kcu.column_name).

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/bin/initdb/initdb.c`:
- Around line 2067-2068: The call to setup_run_file uses a bare filename
"oracle_compat_views.sql" which fails because readfile() expects a full path;
declare a static char *oracle_compat_views_file (near system_views_file), call
set_input(&oracle_compat_views_file, "oracle_compat_views.sql") in
setup_data_file_paths(), add check_input(oracle_compat_views_file) alongside
other checks, and change the setup_run_file call to setup_run_file(cmdfd,
oracle_compat_views_file); also consider wrapping the load of
oracle_compat_views_file in a conditional based on database_mode == DB_ORACLE to
match other Oracle-specific files.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/bin/psql/psqlplusparse.c`:
- Around line 2018-2038: get_guc_settings can return NULL on any failure and
callers pass its return to pg_strcasecmp, causing a NULL deref; change
get_guc_settings to always return an allocated string (never NULL) by setting
val = pg_strdup("") on any non-success path and only pg_strdup the real value
when the PGresult is valid and non-null; also ensure you check that res is
non-NULL and still call PQclear(res) and destroyPQExpBuffer(query) before
returning so callers (e.g., sites that call pg_strcasecmp) always receive a
valid empty string instead of NULL.
- Around line 1732-1737: The code copies the EXEC statement body from the wrong
place by using strlen(state->scanbuf), which points to the end of the outer
buffer; instead capture the remainder of the current line via state->scanline.
In the block that creates psqlplus_cmd_execute (make_exec_node / psqlpluscmd),
replace the offset/substring logic that uses strlen(state->scanbuf) with a
direct copy of the rest of the current line (use pg_strdup(state->scanline) or
otherwise copy from state->scanline) so exec->plisqlstmts contains the text
after EXEC/EXECUTE.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@contrib/ivorysql_ora/sql/ora_sysview.sql`:
- Around line 86-98: Add more cases around TEST_USER_CONS_COLUMNS to exercise
composite and other constraint types: create additional tables (e.g., TEST_REF
for FK) and alter/create TEST_USER_CONS_COLUMNS with multi-column PRIMARY KEY,
UNIQUE constraint, FOREIGN KEY referencing TEST_REF, and a CHECK constraint so
USER_CONS_COLUMNS returns rows for each constraint type; insert sample rows that
exercise column positions for composite constraints and ensure queries select
OWNER, TABLE_NAME, CONSTRAINT_NAME, COLUMN_NAME plus POSITION/ORDINAL_POSITION
(or equivalent) to validate column ordering; adjust the SELECT to ORDER BY
CONSTRAINT_NAME, POSITION (or COLUMN_NAME position field) and add
assertions/expected outputs for each constraint type; finally ensure proper
cleanup by dropping the created tables and constraints (references:
TEST_USER_CONS_COLUMNS, TEST_REF, USER_CONS_COLUMNS, CONSTRAINT_NAME,
POSITION/ORDINAL_POSITION).

In `@contrib/ivorysql_ora/src/sysview/sysview--1.0.sql`:
- Around line 1196-1216: After creating the SYS.USER_CONS_COLUMNS view, add a
GRANT SELECT on that view to PUBLIC so regular users can read it; locate the
CREATE OR REPLACE VIEW SYS.USER_CONS_COLUMNS block and append a GRANT SELECT ...
TO PUBLIC statement immediately after the view definition (consistent with
neighboring SYS views) to ensure the same access semantics as the other SYS.*
views.
- Around line 1196-1215: The SYS.USER_CONS_COLUMNS view is missing the POSITION
column and a GRANT SELECT to PUBLIC, and its ORDER BY should preserve
multi-column constraint order; update the CREATE OR REPLACE VIEW
SYS.USER_CONS_COLUMNS to include a POSITION expression (derived from the array
index of ATT.ATTNUM within CON.CONKEY), add POSITION to the select-list after
COLUMN_NAME (keeping OWNER, TABLE_NAME, CONSTRAINT_NAME, COLUMN_NAME), change
the trailing ORDER BY to ORDER BY TABLE_NAME, CONSTRAINT_NAME, POSITION, and add
a separate GRANT SELECT ON SYS.USER_CONS_COLUMNS TO PUBLIC statement after the
view definition.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@contrib/ivorysql_ora/src/sysview/sysview--1.0.sql`:
- Around line 1218-1241: The file mistakenly redefines the constraint metadata
view SYS.USER_CONS_COLUMNS with a table-column metadata query; revert this by
removing or renaming the CREATE OR REPLACE VIEW SYS.USER_CONS_COLUMNS block so
the original constraint-oriented view (with OWNER, CONSTRAINT_NAME and full
CONSTRAINT_TYPE mapping P/U/F/C derived from constraints, not just pg_index) is
preserved; if the table-column query belongs as a separate catalog object, give
it a distinct view name (not SYS.USER_CONS_COLUMNS) and ensure it does not
shadow OWNER/CONSTRAINT_NAME or overwrite the constraint-type logic that the
real USER_CONS_COLUMNS must provide.