External view of the software

Software files necessary to operate

PyADQL is a self-contained Python library and CLI: it needs no external database, no network access, and no runtime configuration file to operate. The files it depends on at runtime all ship inside the installed package (src/pyadql/, installed as the pyadql distribution):

File / directory	Role
pyadql/grammar/adql.lark	Entry point of the ADQL grammar; assembles the files below via Lark %import
pyadql/grammar/lexer.lark	Terminals: NAME, STRING, numeric literals, keyword tokens
pyadql/grammar/literals.lark	Identifiers, literals, CAST target types
pyadql/grammar/core.lark	Query structure, predicates, expressions, functions, geometry
pyadql/parser.py	Compiles the grammar (lazily, cached) and transforms a Lark parse tree into the typed AST
pyadql/ast_nodes.py	The AST node dataclasses themselves (no dependency on Lark)
pyadql/__main__.py	The pyadql CLI entry point
pyadql/config.py	Loguru logging configuration

The Lark grammar is compiled in memory on first use and cached for the process's lifetime (see parser._get_lark_parser()); PyADQL never writes to disk, and produces no persistent data files, caches, or databases of its own.

Security and privacy considerations

• PyADQL performs no network I/O and reads no external files unless explicitly asked to (pyadql -f PATH, or the query text supplied by the caller). It never executes the ADQL query against a database -- it only parses it -- so there is no SQL-injection surface within PyADQL itself; it can in fact be used to help detect suspicious input before it reaches a real TAP service (see the production-log validation notes in reference_manual.md).
• The only user data PyADQL handles is the ADQL query text passed to it (as a CLI argument, a file, or standard input) and the resulting AST, both of which stay in memory / on the caller's own stdout; nothing is transmitted elsewhere.
• Diagnostic logs (via loguru) go to stderr and, depending on the verbosity level requested (--level), may include the raw query text (DEBUG) -- treat logs at DEBUG/TRACE level as containing the same sensitivity as the queries being parsed.

Continuity of operation

PyADQL is stateless: every call to parse_adql() (or every pyadql invocation) is fully independent, with no session, lock file, or on-disk state to corrupt or recover. In case of an unexpected interruption (e.g. Ctrl+C, trapped by the CLI's SigintHandler), simply re-running the same command resumes normal operation -- there is no cleanup or recovery procedure required.