std.format spec conflict: docs say "follows Python rules" but std.jsonnet rejects booleans for numeric format codes

Start with the std.format documentation at the linked stdlib reference and the format_code helper in std.jsonnet, then compare the behavior described there with the examples in the issue. Review the linked sjsonnet PR for implementation context; done means the intended boolean behavior is decided and the documentation and reference implementation no longer conflict.

Summary

There is a conflict between the official documentation and the reference implementation (std.jsonnet) regarding how std.format handles boolean values with numeric format codes.

Documentation says

From https://jsonnet.org/ref/stdlib.html#std-format:

The string formatting follows the same rules as Python.

In Python, bool is a subclass of int, so all numeric format codes accept booleans:

>>> "%d" % True
'1'
>>> "%f" % True
'1.000000'
>>> "%x" % True
'1'
>>> "%s" % True
'True'

Reference implementation says

In std.jsonnet, the format_code helper explicitly checks:

if std.type(val) != 'number' then
  error 'Format required number at ' + i + ', got ' + std.type(val)

Since std.type(true) returns 'boolean' (not 'number'), all numeric format codes reject booleans:

std.format("%d", true)  → ERROR: "Format required number at 0, got boolean"
std.format("%f", true)  → ERROR
std.format("%x", true)  → ERROR

Only %s works: std.format("%s", true) → "true"

The conflict

This causes confusion for implementors: should a compliant Jsonnet implementation follow the documented Python behavior (accept booleans) or the std.jsonnet implementation (reject booleans)?

Question

Which is the intended behavior?

• (A) The documentation is correct — std.format should follow Python's rules and accept booleans for numeric codes (meaning std.jsonnet has a bug)
• (B) The implementation is correct — std.format should reject booleans for numeric codes (meaning the documentation should be updated to remove the "follows Python" statement)

Context

This was discovered in sjsonnet where we initially implemented option (A) (following Python), then reverted to match std.jsonnet's behavior. See: https://github.com/databricks/sjsonnet/pull/1016