sort_keys does not (always?) work

[lehmannro]

dumps(sort_keys) does not seem to work for parsed TOML documents:

>>> tomlkit.dumps(tomlkit.loads('zzz = 1\naaa = "foo"'), sort_keys=True)
'zzz = 1\naaa = "foo"'

It works fine on regular objects, as far as I can tell:

>>> tomlkit.dumps({'zzz': 1, 'aaa': "foo"}, sort_keys=True)
'aaa = "foo"\nzzz = 1\n'

It does work when I manually apply the item(_sort_keys) wrapping done in the API:

>>> tomlkit.dumps(tomlkit.item(tomlkit.loads('zzz = 1\naaa = "foo"'), _sort_keys=True))
'aaa = "foo"\nzzz = 1\n'

I believe it's because the condition doesn't trigger since tomlkit.loads returns a TOMLDocument:

>>> x = tomlkit.loads('zzz = 1\naaa = "foo"')
>>> type(x)
<class 'tomlkit.toml_document.TOMLDocument'>
>>> isinstance(x, tomlkit.container.Container)
True

[frostming]

Reordering a parsed toml document will cause many format issues, so ideally you should always pass a plain dictionary as the argument of tomlkit.dumps()

[lehmannro]

I'm trying to fix an issue in a formatter which uses a loads/dumps cycle to sort keys, so that's not an option here.

I could pass through a dictionary, but then as you said I'd lose all formatting:

>>> tomlkit.dumps(dict(tomlkit.loads('zzz = 1\n# bar\naaa = "foo"')), sort_keys=True)
'aaa = "foo"\nzzz = 1\n'

I understand there's no single way to perform sorting in that case (e.g., does the # bar comment belong after zzz or before aaa?). Would accessing TOMLDocument.body be the only / a supported way of looking at the items myself?