Wrapped large diff remains blank on v0.17.6 while v0.13.2 renders

[mplibunao]

Summary

Hunk v0.17.6 remains blank indefinitely when rendering a large diff with line wrapping enabled. The same diff renders and remains navigable on v0.13.2, although memory usage is high. Disabling wrapping on v0.17.6 restores sub-second rendering.

This appears distinct from the general macOS startup regression reported in #534: that original small-fixture reproduction no longer regresses on v0.17.6.

Environment

• macOS 26.5.2, arm64
• Kitty 0.45.0
• Fish 4.8.0
• Native arm64 Hunk binaries built from the corresponding tagged Nix flakes
• Hunk config:

agent_notes = true
line_numbers = true
theme = "ember"
watch = false
wrap_lines = true

Reproduction

From an already-running Fish shell in Kitty, in a branch with 141 changed files, 20,852 insertions, and 1,475 deletions:

hunk diff main

Git produces the underlying diff in approximately 0.02 seconds.

Results

Version / configuration	Result
v0.17.6, wrapping enabled	Blank after several minutes; sampled at 2m42s with approximately 1 GB physical footprint and the main thread idle in the event loop
v0.17.6, --no-wrap	Complete diff rendered in 0.63s; navigation responsive; approximately 345 MB RSS during the observed run
v0.17.6, wrapping enabled, scoped to one 648-line file	Rendered in 0.65s
v0.13.2, wrapping enabled	Complete diff rendered in 2.83–2.85s; navigation responsive; approximately 2.28 GB RSS

The comparisons used the same machine, Kitty instance, repository, terminal dimensions, diff, and configuration. Each run used a private HUNK_MCP_PORT. The literal wrapper path hd main only forwards to hunk diff main and reproduced the same results.

Expected behavior

Wrapped reviews should either:

1. render with bounded memory and responsive navigation, or
2. degrade gracefully with progressive or single-file rendering, or a clear size warning, instead of leaving a blank TUI indefinitely.

Relevant existing work

• high memory usage and slow ux when hunking over repo with lots of new files/changes #405 covered high memory usage on large reviews and was closed after the large-review memory and performance work.
• Performance problems on large diffs #413 covered large-diff interaction latency and was closed after v0.15.3.
• Feature Request: Allow persistent default for line wrapping (wrap = true) #560 confirms wrap_lines = true is a supported preference.
• PR perf: reduce large-review scroll and navigation latency by ~100–1000× #427 notes that wrapped mode does not use file-section windowing because wrapped rows have variable heights, leaving wrapped reviews on the full render path.

This may be a regression or an uncovered scalability boundary in that wrapped full-render path.