Stop joined JSON containers from indenting nested sequences

[ESultanik]

Closes #158

What was wrong

SequenceFormatter.print_SequenceNode always wraps its items in self.items_indent(printer), and the default implementation always calls printer.indent(). Printer.write materializes indentation lazily, on the first write after a newline, so a container that never emits a newline still raises the indent counter for everything nested inside it.

--join-dict-items suppresses the newlines between key/value pairs, so the dict's extra level of indentation was invisible at the dict level but still counted. A nested list added its own level and did emit newlines, so its items came out eight spaces deep and its closing bracket four. --join-lists had the same latent problem for anything nested inside a joined list.

JSONListFormatter and JSONDictFormatter now return the printer unchanged from items_indent when their join option is set, so indentation tracks the depth that actually breaks lines. Printer is itself a context manager, and pydiff, yaml, ini, and csv already return one this way.

Before and after

All examples are {"foo": [1, 2, 3], "bar": "baz"} diffed against itself.

graphtage -jd a.json a.json:

{"bar": "baz","foo": [
        1,
        2,
        3
    ]}

{"bar": "baz", "foo": [
    1,
    2,
    3
]}

graphtage -jl a.json a.json was {\n "bar": "baz",\n "foo": [1,2,3]\n} and is now:

{
    "bar": "baz",
    "foo": [1, 2, 3]
}

graphtage -j a.json a.json was {"bar": "baz","foo": [1,2,3]} and is now:

{"bar": "baz", "foo": [1, 2, 3]}

Nesting now composes. With -jd on {"x": [{"y": [1, 2]}, 3]}, only the two lists break lines, so only they indent:

{"x": [
    {"y": [
        1,
        2
    ]},
    3
]}

The delimiter space

Both options now emit a single space in place of the newline they suppress, so joined output reads [1, 2, 3] and {"bar": "baz", "foo": …}. The space is applied in item_newline, which runs only between items, so it never leaves trailing whitespace before a newline in the unjoined case, and it does not appear before the first item or after the last.

Applying it to both options keeps --condensed self-consistent, and it matches the spacing the same documents already get when pretty-printed, where "bar": "baz" and the list items are separated by a comma and a line break rather than a bare comma.

Validation

• New test/test_json.py covers -jl, -jd, and both together, including the nested-list case, the deeper {"x": [{"y": [1, 2]}, 3]} case, the delimiter spacing, and a joined diff containing an insertion and a removal. There was no test for either option before.
• Against the unfixed formatter, 7 of the 8 new tests fail; the eighth is the default pretty-printed baseline, which this change must not alter. All 8 pass with the fix.
• ruff check graphtage test docs bindist: clean.
• pytest: 148 passed, including the test/test_formatting.py fuzz round-trip suite (1000 JSON iterations).
• make -C docs html SPHINXOPTS="-W --keep-going": build succeeded.
• uv lock --check fails in my environment because of a global exclude-newer setting; it fails identically on an unmodified tree, and no dependency metadata is touched here.
• Checked that --html -jd emits a single indentation div for the nested list, and that pydiff output is byte-identical without the join options.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GypKU5KdLfs2Cf8kS2TzJa