SettingsFrame.settings passes a Schema class to ListField's item_type instead of a SchemaField

[JarryShaw]

SettingsFrame.settings passes a Schema class where ListField expects a field instance, with the resulting type error silenced by a # type: ignore[arg-type].

The site

pcapkit/protocols/schema/application/httpv2.py:283-285:

#: Settings.
settings: 'list[SettingPair]' = ListField(
    length=lambda pkt: pkt['__length__'],
    item_type=SettingPair,  # type: ignore[arg-type]
)

Every other ListField in the tree wraps its schema item in a SchemaField — for example pcapkit/protocols/schema/transport/tcp.py:393 is item_type=SchemaField(length=8, schema=SACKBlock), and internet/hip.py:260, internet/mh.py:535 and transport/sctp.py:702 follow the same shape.

Confirmed on e2d8ed6d1

SettingPair is a SchemaMeta -> Schema subclass: True
ListField._item_type is: SchemaMeta

So _item_type holds the metaclass-produced class object rather than a field, and the type: ignore is what let it through review — mypy had it right.

Consequence

ListField.unpack's schema branch calls self._item_type(packet) to build a configured per-item field (see #433, which corrected a related bug on that line). Calling a Schema class with a packet dict does not produce a field; it attempts to construct a schema instance from a dict positionally. Depending on the branch taken, this either raises or silently mis-parses.

Why this was unreachable until now

It sits behind the nested-packet fault in #445: the SETTINGS frame's pack path died earlier with KeyError: 'length'. PR #457 fixes #445, and this becomes reachable — the agent implementing it reported SETTINGS as one of five HTTP/2 frames that then fail on distinct, previously-unreachable defects.

Fix

Wrap it as the siblings do:

item_type=SchemaField(schema=SettingPair),

and remove the # type: ignore[arg-type], confirming with mypy rather than assuming — the baseline is 128 errors in 41 files. A regression test should exercise a real SETTINGS frame round trip rather than a fake, since a fake with a compatible __call__ would hide the distinction.

Provenance

Surfaced by the agent implementing #445 (PR #457) and verified independently before filing. Related: #433 (the same ListField schema branch), and #445 which currently masks this.

[JarryShaw]

Correction to this issue's text: the masking symptom is KeyError: 'flags', not KeyError: 'length'.

I wrote that the SETTINGS pack path "dies earlier with KeyError: 'length'". That is wrong, and PR #462's reviewer caught it. Reproduced on 0283a6d59:

proto = object.__new__(HTTPv2)
schema = proto.make(type=Frame.SETTINGS, sid=0,
                    frame={'settings': [(Setting.HEADER_TABLE_SIZE, 4096)]})
bytes(schema)

File ".../pcapkit/protocols/schema/application/httpv2.py", line 144, in post_process
    if packet['flags'][name]:
KeyError: 'flags'

So the field that cannot be reached is flags, not length, and the raising site is FrameType.post_process at httpv2.py:144 rather than a length callback. 'length' was carried over from #445's CGA Parameters symptom, which genuinely is KeyError: 'length' on a different field in a different module — I reused the phrasing without re-checking what SETTINGS actually raises.

The substance of this issue is unaffected, and worth restating so the correction is not read as a retraction:

• SettingsFrame.settings really does pass the SettingPair class to ListField's item_type where a field instance belongs, with the type error silenced by # type: ignore[arg-type]. ListField._item_type holding a SchemaMeta is still the defect.
• It really is masked by A nested schema cannot reach the enclosing packet's fields by name, so CGA Parameters raises KeyError: 'length' #445 today — just by a flags lookup rather than a length one. Both are the same underlying fault, a nested schema unable to reach an enclosing field by name, so the masking claim holds.
• EXPECTED_FAILURES['httpv2-frame/SETTINGS'] records Gap('CONSTRUCT', "KeyError: 'flags'", …) — the harness had the right symptom all along, which is where I should have checked rather than reusing A nested schema cannot reach the enclosing packet's fields by name, so CGA Parameters raises KeyError: 'length' #445's wording.

Noting for whoever picks this up: the same wrong symptom propagated into PR #462's test docstring and description, and its reviewer has flagged it there too.