scripts/repo_shape.py reads ? a: b in a flow mapping as the literal key ? a, where the block reader refuses explicit keys, and refuses {a: } and {a:} as not a flow collection: '' instead of reading {'a': None} #252

Lane: openreposhape-5 (openRepoShape-5)

Found by this lane's adversarial review of #213's fix as F3 and F4, two edges of the flow-mapping reader that the fix did not touch. Pre-existing on main 54aa69a, and the same on fix/yaml-flow-escaped-quote 86ffe66 (PR #244) and fix/adopt-plan-linebreak-names a02240a (PR #247). Latent: no writer in this repository emits a flow mapping with an explicit key or an empty value (see Reach).

What is wrong, two ways.

1. An explicit key in a flow mapping is read as a key. The block reader refuses ? by name, and the reader's header lists explicit keys as REFUSED, "an invented meaning in a validator is a wrong answer with a confident tone". _refuse_unsupported (scripts/repo_shape.py:418) runs only on block lines (called at :533). _split_key_flow (:322) has no such check, so in a flow mapping ? a: b is the literal key ? a.
2. An empty value in a flow mapping is refused as a flow collection. _flow_value (:299-301) is _flow(text) if text[:1] in "[{" else _scalar(text), and for the empty string ''[:1] in "[{" is True ('' in "[{" is True for any string), so _flow('') runs and raises not a flow collection: ''. The block twin a: reads {'a': None}, and so does YAML.

SHAPE=<openRepoShape checkout>
SHAPE=$SHAPE python3 - <<'PY'
import os, sys
sys.path.insert(0, os.environ["SHAPE"] + "/scripts")
from repo_shape import parse_yaml, YamlError
def read(text):
    try:
        return repr(parse_yaml(text))
    except YamlError as exc:
        return f"YamlError: {exc}"
for text in ("k: {? a: b}\n",            # F3: an explicit key in a flow mapping
             "k:\n  ? a\n  : b\n",       #     the same construct in a block: refused
             "k: {a: }\n",               # F4: an empty value in a flow mapping
             "k: {a:}\n",
             "k: {a: , b: 1}\n",
             "k:\n  a:\n  b: 1\n"):      #     the block twin of F4: a null value reads fine
    print(f"{text!r:24} ->", read(text))
PY

Observed on main 54aa69a (the same on 86ffe66 and on a02240a):

'k: {? a: b}\n'          -> {'k': {'? a': 'b'}}
'k:\n  ? a\n  : b\n'     -> YamlError: line 2: explicit keys are not supported
'k: {a: }\n'             -> YamlError: not a flow collection: ''
'k: {a:}\n'              -> YamlError: not a flow collection: ''
'k: {a: , b: 1}\n'       -> YamlError: not a flow collection: ''
'k:\n  a:\n  b: 1\n'     -> {'k': {'a': None, 'b': 1}}

PyYAML 6.0.3 reads k: {? a: b} as {'k': {'a': 'b'}} and each of the three empty-value rows with 'a': None ({'a': None, 'b': 1} for the third). Row 1 is a wrong answer with no refusal: the reader's own rule, applied in a block, would have raised. Rows 3 to 5 are refusals of valid YAML, with a message that names an empty string a person never wrote.

Expected. ? a: b in a flow mapping is refused as the block form is, explicit keys are not supported, since the reader declines to invent a meaning for it; the refusal sits where both forms meet, not only on block lines. {a: }, {a:} and {a: , b: 1} read {'a': None} and {'a': None, 'b': 1}: _flow_value treats an empty value as null (_scalar('') already returns None), and the [{ test asks whether the text OPENS a flow collection (text[:1] in ("[", "{"), or text.startswith(("[", "{"))), which an empty string does not. A test pins the rows above.

Reach. Nothing writes either form. The four flow writers emit sequences of bare words (adopt-project.py pins and allow_empty_legs, scripts/shape_materialize.py also_matches and referent_chain) and members: [] / paths: []; no writer emits a flow mapping with a key or value. So it shows only on a file a person edits or another tool writes, a flow mapping such as naming: {form: } left half-filled in project.yaml. None of the 14 tracked YAML files holds a flow mapping other than a template placeholder (git ls-files '*.yaml' '*.yml', searched for : {, - { and a line opening ? ).

Adjacent and not part of this issue, from the same review: empty flow ITEMS are dropped silently ([,] reads [], [a, , b] reads ['a', 'b'], where YAML refuses the second) and [k: v] reads as the one string k: v; both are older and untouched by the fix. They are listed so the reader's next differential check covers them.

Found by. handoffs/openRepoShape/attachments/lane-openRepoShape-5/review213-report.md in opensoft/brett-wip (F3 and F4). Re-run on main 54aa69a, 86ffe66 and a02240a for this issue.

Not claimed.

🤖 Generated with Claude Code