[tae898]

Refs #267 (row 8). Adds the page how-to/functions-and-events.adoc (Python functions in SQL and openCypher, and db.on record events), one nav line, and links from the quickstart, installation, databases-and-queries and transactions pages (one added anchor, failed-statement, on the transactions page).

The existing tests (test_python_callbacks.py, test_native_udf.py, test_python_function_interrupt.py) already pin the basics, so the page header links to them and tests/test_functions_and_events_behavior.py pins only what the page adds: the Python types a function receives from a stored value (datetime, date, Decimal, list, map, and a vertex, edge or document as its wrapper), openCypher arguments and results, that a function runs once per row, that a raising function or callback undoes its statement and leaves the transaction going (26.11.1), that a function belongs to the process and an event to one database (and what closing a database drops), which thread runs a function or an event, that a waiting function blocks only its own statement, when an after_* event fires relative to the commit, only False vetoes, the exact-type filter, and what a callback may do (read, write into the same transaction, fire further events, change the record). tests/test_docs_functions_page.py runs every Python block of the page in order and checks the printed output.

CI: the first JVM run found one wrong page statement (get_type_name() is a Java string on arcadedb-jvm; see the surprises); it is fixed, and the JVM test workflow on that head passed (24 of 24 jobs). The head was then rebased onto main after #282 and its run is in progress; the native parity legs run now that the PR is ready. I could not run pytest on the laptop (a guard blocks it), so the native-side expectations are first observed by CI. Issue #286 tracks the get_type_name() inconsistency; a fix PR follows after this one merges.

Surprises, none filed:

• A before_create callback that calls record.set(...) changes what is stored on arcadedb-jvm (the callback holds the engine's mutable record) and has no effect on arcadedb (it holds a snapshot dict built in NativeExtras.appendRecord, and only the veto flag goes back). The test pins both and the page says not to rely on it.
• On arcadedb, a before_update callback gets a record with an RID, which the native wrapper treats as read-only, so record.set(...) there raises instead of being ignored (read from _native/records.py, not tested).
• register_function says the GIL serializes the calls and fn does not run in parallel. That holds only while fn holds the GIL: a function that waits (sleep, Event, I/O) lets other statements run, and a free-threaded build has no GIL. The page says to write fn as if two threads could call it at once.
• record.get_type_name() returns a Java string on arcadedb-jvm (graph.py returns getTypeName() unconverted) and a Python str on arcadedb. It compares equal to a str, but "created " + record.get_type_name() raises TypeError on the JVM build only; the first JVM run caught it in a page block. record.get(...), to_dict() values and get_rid() are str on both. The page says to use str(record.get_type_name()), and a test pins the type per backend, inside a callback and outside one.

What remains of #267 after this: rows 4 and 5 (PR #282), row 7, rows 10 and 11 (another agent), and 1, 6, 13 and the stale items under #264.

🤖 Generated with Claude Code