Nested arrays fail to round-trip: encoder emits a length-1 list-array header over an empty body ([[[]]])

[antrixy]

Description

toon-python (0.9.0b1) mishandles nested array-of-array values in both directions. The reference implementation @toon-format/toon (v2.3.0) round-trips [[[]]] correctly, so both facets are specific to toon-python.

Encode: encode([[[]]]) produces a list-array header ([1]:) that claims one item over a body that yields none, so it cannot be decoded — even by toon-python's own decoder. encode([[[]]]) produces:

[1]:
  [1]:
    - [0]:

decode() of that raises:

File ".../toon_format/decoder.py", line 786, in decode_list_array
    raise ToonDecodeError(f"Expected {expected_length} items, but got {len(result)}")
toon_format.decoder.ToonDecodeError: Expected 1 items, but got 0

Decode: decoding the reference (TypeScript) encoding of [[[]]] returns an object instead of an array — the internal list-array markers leak out as keys:

[{"[1]": {}, "-": []}]

The decoder facet is not specific to empty arrays; any nesting depth >= 3 reproduces it (e.g. the reference encoding of [[{"s":"a:b"}]] decodes to [{"[1]": {}, "- s": "a:b"}]).

Boundary (reduced with a differential fuzzer + delta-shrinker):
[] — fine
[[]] — fine
[[[]]] — breaks (minimal reproducer)

Environment:
toon-python (toon_format): 0.9.0b1
reference impl @toon-format/toon: v2.3.0 (round-trips this input correctly)
Python: 3.14

Reproduction Steps

1. from toon_format import encode, decode
2. Call encode([[[]]]) — produces:
   [1]:
   [1]:
   - [0]:
3. Call decode() on that output — raises ToonDecodeError: Expected 1 items, but got 0
   (toon-python cannot decode its own encoding)

Separately, decoding the reference @toon-format/toon encoding of [[[]]] returns
[{"[1]": {}, "-": []}] — a nested array decoded as an object.

Expected Behavior

[[[]]] round-trips unchanged: decode(encode([[[]]])) == [[[]]].
(The reference impl @toon-format/toon does this correctly.)

Actual Behavior

encode([[[]]]) produces:

[1]:
  [1]:
    - [0]:

The outer header says [1]: (one item) but the innermost level is [0]: (zero items).
decode() of that raises:

File ".../toon_format/decoder.py", line 786, in decode_list_array
    raise ToonDecodeError(f"Expected {expected_length} items, but got {len(result)}")
toon_format.decoder.ToonDecodeError: Expected 1 items, but got 0

Separately, decoding the reference @toon-format/toon encoding of [[[]]] returns an
object instead of an array:

[{"[1]": {}, "-": []}]

Environment

toon-python (toon_format): 0.9.0b1
reference impl @toon-format/toon: v2.3.0 (round-trips this input correctly)
Python: 3.14

Additional Context

The decoder facet is not specific to empty arrays; any nesting depth >= 3 reproduces it
(e.g. the reference encoding of [[{"s":"a:b"}]] decodes to [{"[1]": {}, "- s": "a:b"}]).

Boundary, reduced with a differential fuzzer + delta-shrinker:
[] — fine
[[]] — fine
[[[]]] — breaks (minimal reproducer)

[antrixy]

Corroborating this from a differential fuzz sweep — the encode facet reproduces on non-empty nested arrays too, so it's about nesting depth, not empty bodies.

encode([[[1]]]) produces:

[1]:
  [1]:
    - [1]: 1

Same defect as the [[[]]] repro above: the middle [1]: is emitted as a bare list-array header with no - marker, so it isn't a valid list item. All three decoders reject it, including toon-python's own (Expected 1 items, but got 0) — identical failure surface to the empty-body case. The boundary matches the one in the description exactly: [[1]] fine, [[[1]]] breaks.

New characterization — the bug is specific to the homogeneous arrays-of-arrays path. A mixed array whose elements aren't all arrays takes the general list path and encodes correctly. encode([1, [[2]]]) produces:

[2]:
  - 1
  - [1]:
    - [1]: 2

Here the nested-array element gets its - [1]: marker correctly (line 3). So the marker omission only happens when every element of the parent array is itself an array — that branch skips the list-item prefix that the mixed-element branch emits. That localizes the encoder fix to the all-arrays encoding branch.

Reference (@toon-format/toon v2.3.0) encodes [[[1]]] as [1]: / - [1]: / - [1]: 1 — - marker at every level — and round-trips it correctly, consistent with the reference behavior already noted for [[[]]].