Upgrade to cel-rust 0.14, extended stdlib, and prepare 0.7.0

[hardbyte]

Summary

Prepares a 0.7.0 release: upgrades to cel-rust 0.14, ships an
opt-in extended standard library that mirrors cel-go, and exposes
expression static analysis. Fully backward compatible for the common API
(evaluate, compile, Context) apart from the documented cel 0.14 behaviour
changes.

Draft for maintainer review before release. A 1.0 release is deferred until the upstream cel crate supports the protobuf AST (cel.expr.Expr / CheckedExpr), enabling cross-implementation portability.

What changed

Upgrade to cel-rust 0.14 (src/lib.rs)

• Bumped cel 0.13 → 0.14.
• Shared standard-library Env built once (LazyLock<Arc<Env>> + Context::with_env) instead of rebuilt on every call — a small perf win for repeated evaluate()/Program.execute().
• Reworked error mapping for the 0.14 ExecutionError variants: canonical CEL type names in messages (Unsupported operation: string + int) and more variants mapped to idiomatic Python exceptions (UnsupportedIndex, ValuesNotComparable, UnexpectedType, InvalidArgumentCount → TypeError).
• add_variable_from_value (infallible) replaces the fallible add_variable call.

Custom functions callable as methods (src/lib.rs)

The custom-function bridge now reads the FunctionContext directly and prepends the method receiver, so a registered function works as both f(x, y) and x.f(y). This is what lets the standard-library extensions be written as members ("s".charAt(i), [1,2].contains(x)).

Extended standard library (python/cel/stdlib.py)

New opt-in libraries mirroring cel-go: core (bool/dyn/type/min/max), strings, math, sets, encoders (base64), lists. Enable with add_stdlib_to_context(ctx) or extensions=[...]; the CLI enables them automatically.

Expression static analysis (src/lib.rs)

Program.variables(), .functions(), .references() and the .source property report what an expression uses without evaluating it.

Behaviour changes (cel 0.14)

• contains is string-only; use in for list/map membership (or the lists extension).
• min/max moved out of core into the core stdlib extension.
• Built-in functions take precedence over same-named user functions.

Portability / "bytecode"

Out of scope and documented as such: CEL has no bytecode; ecosystem interchange uses the protobuf AST (cel.expr.Expr/CheckedExpr), which upstream cel-rust cannot yet produce/consume (no protobuf, no type checker). The portable artifact is the CEL source string; use Program.references() for analysis.

Tests

455 passing (was 380). New: test_stdlib_extensions.py, test_introspection.py, test_cel_014_behaviour.py. cargo fmt/clippy -D warnings clean; ruff/mypy clean; doc code blocks execute.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 50bb9ccd31

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Return -1 for overlarge search offsets

When the strings extension is enabled, calls like 'hello mellow'.indexOf('ello', 20) should behave like cel-go and return -1 because the search starts past the end and finds no match; this check instead raises, and lastIndexOf has the same positive-offset problem. That turns valid CEL expressions with a too-large user-supplied offset into runtime errors rather than a no-match result.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Raise on int64-min math.abs

For math.abs(-9223372036854775808), cel-go defines an overflow error because the absolute value cannot be represented as a signed CEL int, but Python's abs() returns 9223372036854775808; the bridge will then convert that to a CEL uint and the expression succeeds with the wrong type/value instead of failing.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve NaN in math.sign

For a double NaN input, both comparisons are false, so math.sign(0.0 / 0.0) falls through to 0.0. The cel-go math extension specifies that math.sign(NaN) returns NaN, so expressions that use math.isNaN(math.sign(x)) to propagate invalid numeric data will silently see zero instead.

Useful? React with 👍 / 👎.