[compiler-segfault]

What's wrong

check_sql_read resolves the dangerous-function denylist through get_dangerous_functions(ds.type), and ds.type for a PostgreSQL datasource is the DB enum value pg (backend/apps/db/constant.py:29). The table itself was keyed postgresql, so the lookup never matched and the whole PostgreSQL entry was dead:

get_dangerous_functions("pg")          # {'version', 'current_user', 'user', 'database'}
get_dangerous_functions("postgresql")  # + pg_read_file, pg_write_file, lo_import, lo_export

SELECT pg_read_file('/etc/passwd') therefore passed the read-only guard on a PostgreSQL datasource, as reported in #1438.

Fix

Key the entry by the datasource type that is actually used, and note in the table that keys must be DB.type values so the next dialect cannot drift the same way.

While fixing the mapping, the entry also picks up the rest of the server-side file readers that belong to the same family — pg_read_binary_file, pg_ls_dir, pg_stat_file — which were never listed at all. Drop that part of the hunk if you would rather keep this patch to the reported four.

No behaviour change for any other dialect.

Verification

New backend/tests/test_postgres_denylist.py drives the real guard rather than the table alone:

• the seven PostgreSQL file functions resolve and are rejected by check_sql_read with a dangerous function reason,
• INSERT / UPDATE / DELETE / TRUNCATE / COPY / DROP stay rejected on PostgreSQL,
• an ordinary SELECT is still allowed,
• every key in DS_SPECIFIC_DANGEROUS_FUNCTIONS is a DB.type value, so a key mismatch cannot silently disable a dialect again,
• the SQL Server / MySQL entries are unchanged and do not leak across dialects.

pytest backend/tests/test_postgres_denylist.py   17 passed
ruff check / ruff format --check                 clean

Same suite against the unpatched file: 9 failed, 8 passed.

Fixes #1438