Clarify that format strings accept arbitrary tokens as field names

[lgalfaso]

Bug report

Bug description:

Should the code below produce an error?

'{(}'.format(**{'(': 2})

This produces '2' in Python 3.14.5 on MacOS.

Reading
https://docs.python.org/3/library/string.html#format-string-syntax

replacement_field: "{" [field_name] ["!" conversion] [":" format_spec] "}"
field_name:        arg_name ("." attribute_name | "[" element_index "]")*
arg_name:          [identifier | digit+]

this ends up pointing to identifier at https://docs.python.org/3/reference/lexical_analysis.html#grammar-token-python-grammar-identifier

NAME:          name_start name_continue*
name_start:    "a"..."z" | "A"..."Z" | "_" | <non-ASCII character>
name_continue: name_start | "0"..."9"
identifier:    <NAME, except keywords>

Given that '(' is an Ascii character, then ( is not a valid identifier and it should not be allowed.

CPython versions tested on:

3.14

Operating systems tested on:

macOS

Linked PRs

• gh-150626: Fix format string syntax grammar #151064
• gh-150626: Update Format String Syntax documentation #153143

[picnixz]

I think it would be too annoying to actually change the parser and it doesn't benefit us to restrict this either. Instead, we can update the docs to make argname allow any token. Note that:

>>> '{(.__class__}'.format(**{'(': int})
"<class 'type'>"

So even the ("." attribute_name | "[" element_index "]")* part is correct. I would still suggest doing a bit more tests to see if the arg_name can be replaced by any token.

[lgalfaso]

A quick check shows that the parser allows for any character other than '!', '.', ':', '[', '{', '}'.

for i in range(0x110000):
    try:
        c = chr(i)
        assert(('{' + c + '}').format(**{c: 2}) == '2')
    except Exception:
        if not c.isdigit(): print("%x -> %c" % (i, c))

21 -> !
2e -> .
3a -> :
5b -> [
7b -> {
7d -> }

Agree that the best way forward would be to update the documentation.

[picnixz]

Thanks for checking! I will update the docs tomorrow.

[robsdedude]

Note also that the grammar's

attribute_name:    identifier

is inaccurate as shown by

class C: ...

c = C()
setattr(c, "(", "hi")
'{(.(}'.format(**{'(': c})  # -> 'hi'

EDIT
I adjusted the test from above and it appears the attribute_name follows the same rules as the field_name.

class C: ...

c = C()

for i in range(0x110000):
    try:
        attr_name = chr(i)
        c.__dict__ = {attr_name: 'hi'}
        assert(('{c.' + attr_name + '}').format(c=c) == 'hi')
    except Exception:
        print("%x -> %c" % (i, attr_name))

yields

21 -> !
2e -> .
3a -> :
5b -> [
7b -> {
7d -> }

I've also checked with try that neither block keywords.

c.__dict__ = {"try": 'hi'}
print('{try.try}'.format(**{'try': c}))  # -> 'hi'