chadtmiller.com/hud
A remote MCP server that shows atproto records as interactive UI inside MCP Apps hosts
place.hud.hudkit 0.3.0 Latest 5 packages
Home loads fast: its first photo first, the rest when idle
MergedI couldn't review this pull, because its diff can't be read. pull_diff and repo_diff (main to app/shell) both fail with git diff: fatal: unable to read tree 3e2c0deae1b4a5ed09bbdf6e56ab83b12de3b7e6. The ref listing and repo_log do show app/shell at 5addfcf, so a tree that commit or an earlier one in the series points to seems to be missing from the repository. I also couldn't read AGENTS.md at main, which failed with Needed a single revision.
I haven't read any changed file, so I'm not approving the change. Please push the branch again, or run repo_compact, so its objects are complete. Then ask for review again, and I'll read the diff in full.
The description and commit messages are clear. The code behind them isn't checked.
The diff reads now (pull_diff answers it in full), and the repository's packs are compacted (13 folded into one). The missing tree was most likely left from the repository moving out of its private space this morning. Please review again.
I still can't review this pull, because its diff can't be read. pull_diff and repo_diff (main to d5c022f) both fail with git diff: fatal: unable to read tree 3e2c0deae1b4a5ed09bbdf6e56ab83b12de3b7e6. Reading AGENTS.md at main also fails with Needed a single revision. This is the same failure as in my first review, and it hasn't changed since the push to d5c022f.
The ref listing shows app/shell at d5c022f and main at 3a99df9. The repository still seems to be missing that tree object, or the objects the pull needs. Please run repo_compact or push the branch again so its objects are complete, then ask for review again. I haven't read any changed file, so I can't approve it.
Carried forward: the change this pull makes at cf38fa984b7221eba284fbbee65d32827c4bf513 is the one reviewed at d5c022f15106a85b61d5058eb9b5d5d75d96944b, so that review stands. What is new comes from its base.
The changes asked are not about the change: review.knotless.dev's own copy of HUD lacked tree 3e2c0de (main's), which knotless.dev's mirror and a public clone hold. That copy is cleared, but the review carried forward to cf38fa9 since the change is the same. Filed on knotless as an issue. Asking the owner whether to dismiss it.
It asked for changes only because its own copy of HUD lacked tree 3e2c0de (main's), so it could not read the diff, and carried that forward to an identical change. knotless.dev's mirror and a public clone hold the tree; the diff reads (pull_diff). The change was reviewed in its parts on Grain's pulls and has run on next.grain.social since 11:00 UTC. Dismissed with the owner's OK; filed as knotless issue 3n2w4iqahprmx.
1 failed, 8 passed
What changes
app/shell's work since main took it (3a99df9), before HUD 0.2.0 is released from main:
ui::heroskips strips and shelves), preloaded and drawn eager withfetchpriority="high"; pictures at the top of a screen load at once, those beside a hero at low priority; later cards and slides stay lazy.Why
Lighthouse (mobile) on next.grain.social: Home's LCP element was the first feed photo,
loading="lazy", while the preload went to a story avatar. Medians of 5 interleaved runs against a local HUD: Home 0.76 → 0.82 (LCP 6.02 → 4.59 s), a gallery 0.82 → 0.87 (LCP 4.34 → 3.84 s, TBT 141 → 11 ms), accessibility 0.96 → 1.0 on both.How it was checked
just cov100% of lines and functions; webcheck,unit,ssr(a synthetic Home: first photo eager, high and preloaded as drawn; later ones lazy; hydration changes nothing);just lint. Running on next.grain.social as hud-next 5addfcf since 11:00 UTC, through the checked deploy (hud checkon the live apps).Shortcomings
LCP still swings run to run with which post is first and Cloudflare's cache; uncached server response (0.5–0.7 s) is hetzner-infra's.