Skip to content

UI polish, functionality fixes, and README overhaul - #3

Merged
ssalentin merged 6 commits into
mainfrom
readme-release
Oct 14, 2026
Merged

ssalentin merged 6 commits into
mainfrom
readme-release

Conversation

@ssalentin

Copy link
Copy Markdown
Owner

What a user gets

Before: dap-tui worked, but had a few rough functional edges (mount-label
matching missed a case-variant path; a couple of UI width-budget bugs), a
bare-bones README (keybindings table only, no real screenshots, no badges),
and no CI.

After:

  • A real design pass on the Tokyo Night TUI: Dashboard card dead-space and
    narrow-width text collisions fixed, SYNC STATUS footer drops whole fields
    instead of clipping mid-word on narrow terminals, device-card firmware tag
    no longer clips on double-width (CJK) names.
  • Mount-label matching in the alternate-parent fallback is now
    case-insensitive, matching the primary path's behavior.
  • README overhaul: CI/license/version badges, a real Manual section beyond
    the keybindings table, and 2 real rendered-terminal screenshots (asciinema
    • agg against the actual release binary — not ASCII dumps).
  • A new minimal GitHub Actions CI workflow (cargo test on push/PR) backing
    the CI badge.

How it was verified

Mission dap-tui-ui-polish (OPR.99.0.1), 3 slices, each independently
code-reviewed (dev-review@workshop) and QA'd (dev-qa@workshop) in isolated
worktrees, with proofs under
missions/dap-tui-ui-polish/slices/*/proof/ in the OpenRig workspace:

  • Slice 01 (UX/design pass), tip 6092212 on ui-polish-pass: review found
    1 must-fix (SYNC STATUS footer still mid-word-clipping "never" at 80x24,
    contradicting the commit's own claim) — fixed, re-reviewed clean. QA PASS:
    full suite green, fix independently confirmed via live ASCII renders at
    40x15/60x12/80x24/120x34.
  • Slice 02 (functionality fixes), tip 9c9d9a5 on functionality-fixes:
    review clean (cargo test 34/34 + clippy on the changed file, regression
    test proven load-bearing by reverting the fix locally). QA PASS,
    independently re-verified the same way.
  • Slice 03 (README overhaul), tip f953a55 on readme-release: review
    cross-checked the new Manual section's behavioral claims against the real
    code (not taken on trust) — clean. QA caught a real stacking gap: the
    branch was built from slice 01 alone, but the README also described slice
    02's case-insensitive mount-matching behavior, which wasn't actually on
    this branch yet. Fixed by merging the already-approved functionality-fixes
    branch in (f953a55) — no new unreviewed code, just bringing the branch's
    content in line with what it claims.
  • I independently re-ran cargo test on the final branch tip (f953a55)
    myself: 34 passed, 0 failed, 1 ignored (dev helper). I also opened both
    embedded screenshots (docs/screenshots/dashboard.png,
    docs/screenshots/sync.png) and confirmed they're real rendered-terminal
    captures of the actual TUI, not text dumps.
  • Lines we keep: all commits are under the person's own git identity
    (Sebastian Salentin s.salentin@gmail.com), which they set directly
    mid-session — the builder correctly stopped and asked rather than
    inferring it from existing repo history.

Build branch readme-release (tip f953a55) is the superset: it contains
slice 01 + slice 02 (merged in) + slice 03's README work, fast-forwardable
from current main (dcc778c).

What depends on it

No external arteries — this is a standalone TUI tool. The new CI workflow
(.github/workflows/ci.yml) will start running cargo test on this repo's
future pushes/PRs once this merges; nothing currently depends on it running
green yet, since it hasn't executed on GitHub before this PR exists.

Unsure / flagged during the work

  • Screenshot tooling (asciinema + agg + a specific musl build + a font
    package) isn't something this environment had pre-installed; the builder
    installed it mid-mission. Worth confirming that's acceptable sandbox
    behavior going forward (see mission NOTES.md for the full writeup) —
    flagging for visibility, not blocking.
  • No functional regressions found beyond what's listed above; the builder
    used judgment on which "rough edges" to close per the open-ended brief
    and didn't find anything requiring escalation.

🤖 Generated with Claude Code

Dashboard's DEVICE/PLAYER SETTINGS cards stretched to fill the terminal
via Constraint::Min(8), leaving most of the card empty below a handful
of lines on normal/tall terminals. Capped them to their natural content
height instead, and gave the freed space to LIBRARY/SYNC STATUS, which
now vertically centre their content instead of sitting glued to the top.

Along the way this fixed two real clipping defects at 80x24: LIBRARY's
"output" path line and SYNC STATUS's entire ok/err/never + last-sync
summary (the "state at a glance" the card exists for) were silently cut
off below the card's bottom border. Trimmed gratuitous blank separator
lines and merged the summary into one line (with a narrow-width
fallback) to fit.

Also fixed the title bar and status bar running their two halves
together with no separating space on narrow terminals (padding
calculation saturated to 0); both now drop their lower-priority half
(the title's decorative breadcrumb chain, the status bar's device
count/version) before colliding.

Verified: cargo test (33 passed/1 ignored), the dump_ui TestBackend dump
at 40x15/60x12/80x24/120x34, and real rendered-terminal screenshots via
asciinema+agg.
…llback

disk.rs::mounted_by_label (the fallback used when a device is mounted
under a different parent than the configured mount_point, e.g.
/run/media/$USER/$LABEL instead of /media/$USER/$LABEL) matched labels
case-sensitively, even though its own doc comment says it mirrors
udisks::resolve_block_device's label lookup — which has matched
case-insensitively since it was first introduced. udev/vfat labels are
commonly all-uppercase (IPOD) regardless of how a configured mount_point
directory happens to be cased, so the two lookups could disagree about
whether a device is mounted.

Extracted the matching logic into pick_by_label() so it is unit-testable
without a real mount table, and added a regression test.
Review finding on 5336e56: the narrow-width fallback for the ok/err/never
summary checked the full footer string against the card width, but never
re-checked the fallback string itself — so at 80x24 with 5 devices it
still landed in truncate(), char-clipping "never" to "nev…" mid-word,
the exact defect the previous commit's message claimed to fix.

Replaced the single fallback with an ordered list of whole-field
candidates (full summary -> drop "last ..." -> drop the never count ->
drop the err count -> ok count alone), picking the first that fits by
display-cell width (unicode_width, not char count, per review's
non-blocking suggestion — relative_ago is ASCII-only today but this
closes the same bug class dcc778c fixed elsewhere).
Review nits on c6eca46: mounted_by_label's own doc comment still said
"equals label" after that commit made the match case-insensitive, and
pick_by_label's comment claimed to mirror udisks::resolve_block_device's
to_lowercase() when it actually uses eq_ignore_ascii_case (ASCII-only
case-fold vs full Unicode) — behaviorally identical for the FAT32/vfat
label charset this code actually sees, but not a literal mirror as
worded. Both corrected.
- Add a minimal CI workflow (cargo test on push/PR) to back a build
  status badge — this repo had none yet. Also added license (reads the
  existing LICENSE via shields.io's GitHub license badge) and version
  (from Cargo.toml) badges. No badge added without something real behind
  it; the CI badge will go green once this workflow exists on the
  default branch (it 404s pre-merge, which is expected).
- New "Manual" section: prose walkthrough of the four real interaction
  flows (sync a device, mount/unmount, abort, follow logs) beyond the
  terse keybindings table.
- Two real rendered-terminal screenshots (docs/screenshots/*.png),
  captured via asciinema + agg against the actual release binary with a
  throwaway local config (--skip-mirror, no network) — not text/ASCII
  dumps. Demo data includes a genuine tmpfs mount (real kernel mount
  table entry, unmounted after capture) and a hand-written
  sync-state.toml using the app's own real state-file format, so the
  Dashboard/Sync tabs render their actual mounted/synced-history code
  paths rather than an all-empty first-run state.
dev-qa caught that this branch's README claims case-insensitive mount-
label matching (slice 02's fix) while actually branched from slice 01's
tip, not slice 02's — src/disk.rs here still had the old case-sensitive
comparison. Merging slice 02 (already review+QA approved) so the code
backs what the README says.
@ssalentin
ssalentin merged commit 86e8dcb into main Oct 14, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant