Large files on an aindrive drive open several times faster #3

2026-10-12T14:23:26Z

RaymondCho commented

2026-10-12 14:23:26 +00:00

Memona reads a remote file one 1 MiB slice per call, and each slice fetched the whole file again. The last file read is now kept for the slices that follow, reused only while a fresh stat shows the same revision. Follows lead/memona-plugin-aindrive#2

Verified: 40 unit tests; through Memona's host an 8 MiB file read in 1.1 s instead of 6.8 s, and in the app its download took 1.5 s, matching its SHA-256. Nothing on screen changes.

🤖 Generated with Claude Code

Memona reads a remote file one 1 MiB slice per call, and each slice fetched the whole file again. The last file read is now kept for the slices that follow, reused only while a fresh stat shows the same revision. Follows https://git.capsulizers.com/lead/memona-plugin-aindrive/pulls/2 Verified: 40 unit tests; through Memona's host an 8 MiB file read in 1.1 s instead of 6.8 s, and in the app its download took 1.5 s, matching its SHA-256. Nothing on screen changes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

RaymondCho requested review from bot 2026-10-12 14:28:21 +00:00

bot approved these changes 2026-10-12 14:35:07 +00:00

bot left a comment

✅ Approved
Stat-validated cache reuse is correct; no blocking findings.

Review

Strengths

The cache is reset on connection lifecycle changes and keys reuse by path and freshly statted revision.
The regression test covers repeated slices and invalidation after a revision change.

Checks

cargo fmt --check
cargo clippy --locked --all-targets -- -D warnings
cargo test --locked --lib (40 passed)
cargo build --locked --release --target wasm32-wasip2
git diff --check cc320ee..969e1d5

Declined to judge

None.

Profile: codex · gpt-5.6-terra

✅ Approved Stat-validated cache reuse is correct; no blocking findings. ## Review <details> <summary>Strengths</summary> - The cache is reset on connection lifecycle changes and keys reuse by path and freshly statted revision. - The regression test covers repeated slices and invalidation after a revision change. </details> <details> <summary>Checks</summary> - `cargo fmt --check` - `cargo clippy --locked --all-targets -- -D warnings` - `cargo test --locked --lib` (40 passed) - `cargo build --locked --release --target wasm32-wasip2` - `git diff --check cc320ee..969e1d5` </details> <details> <summary>Declined to judge</summary> - None. </details> > Profile: `codex` · `gpt-5.6-terra`

RaymondCho changed target branch from save-and-delete to main

2026-10-13 21:54:32 +00:00

RaymondCho added 1 commit 2026-10-13 21:54:32 +00:00

Large files on an aindrive drive open several times faster

CI / checks (pull_request) Successful in 37s

Details

8d206bac1b

The host reads a file one 1 MiB slice per call, and every slice fetched
the whole file again, so an 8 MiB file crossed the network eight times.
The last file read is now kept for the slices that follow: each slice
still stats the file first and reuses the reply only at the same
revision, so a file changed in between is fetched afresh.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

RaymondCho closed this pull request

2026-10-13 21:54:33 +00:00

RaymondCho reopened this pull request

2026-10-13 21:54:33 +00:00

RaymondCho merged commit 018b7aa25d into main

2026-10-13 21:55:46 +00:00

RaymondCho referenced this pull request

2026-10-15 14:15:01 +00:00

The README shows how a connected drive looks in Memona #4

Sign in to join this conversation.