The local text read path opened the same file for every consumer. A ranged
read of a file within the snapshot cap cost four opens and three decodes:
an 8KiB binary sniff, a streaming scan for the rendered window, a whole-file
read for bracket context, and another whole-file read to hash the snapshot.
Whole-file reads under the structural summarizer paid a fifth. Two of those
readers also ran normalizeToLF over the same bytes.
Read the bytes once at or below SNAPSHOT_MAX_BYTES and derive every view
from them: sniff the leading 8KiB of the buffer, slice the rendered window
out of it under the identical line and byte budgets, index bracket context
into its addressable lines, and hand the normalized text to the snapshot
store and the summarizer. Past the cap nothing wants the whole file, so the
streaming reader stays.
Line byte lengths are walked out of the buffer rather than measured on the
decoded strings, so reported byte counts and the truncation boundary stay
exact for content that is not valid UTF-8.
The buffered text is BOM-stripped for hashing, matching the decoder the
patcher's live read uses. A whole-file read of a BOM file previously hashed
its tag from BOM-bearing text, so the following edit only applied through
stale-hash recovery and told the model the file had changed externally when
it had not.
Also stop rejoining lines into a fresh whole-file string on the way to
tree-sitter when the caller still holds that text, and drop the unread
selectedBytesTotal accounting from the streaming reader.
Measured on 966 differential cases across CRLF, BOM, lone-CR, invalid-UTF-8,
oversized-line, empty, no-trailing-newline, multi-range and raw shapes: the
spurious recovery warning is the only behavioral difference. Raw reads, which
skip the tree-sitter parse that dominates everything else, get 30-45% faster
(2.7MB: 10.3ms -> 5.7ms); non-raw reads 1-3%.
Rolled partial file appends back to their pre-write size and marked malformed resumed sessions for an atomic rewrite.
Retried transient persistence failures from in-memory state and surfaced the first failure in the interactive TUI.
Fixes#8596
- Replaced time-based sleeps and polling loops with event-driven promise resolvers and fake timers across agent and tool tests.
- Migrated test suites to share in-memory auth storage and fixtures using lifecycle hooks.
- Updated catalog model definitions, metadata, and configurations.
Implemented 2025-11-25 Streamable HTTP polling semantics for POST SSE responses: retain event IDs and retry intervals, wait as instructed, and reconnect with GET plus Last-Event-ID until the originating JSON-RPC response arrives.
Extended the shared SSE parser to expose valid id/retry fields and control-only events so reconnecting consumers do not need to reparse raw lines.
Fixes#8264
- Define a centralized `USER_AGENT` constant in `@oh-my-pi/pi-utils` formatted as `omp/<version>`.
- Replace hardcoded and platform-specific user agent strings across AI providers, catalog scrapers, tools, and search providers with the unified `USER_AGENT`.
- Add unit tests for update-cli binary release distribution gating.
Long write/edit streams made the TUI stutter or freeze for seconds at a
time (ui.loop-blocked warnings, keystrokes starving while the single JS
thread rebuilt previews). Two compounding quadratic paths:
1. parseStreamingJsonThrottled re-parsed the entire accumulated args
buffer on a FIXED 256-byte cadence. The comment claimed this bounded
mid-stream work to O(N), but a constant growth gate still parses an
N-byte buffer N/256 times at O(N) each: O(N^2) with a smaller
constant. The gate now scales geometrically (max(256, len/32)), so
parse points form a geometric progression: O(log N) parses, O(N log N)
total, with mid-stream snapshots staying within ~3% of the stream.
Small buffers keep the exact fixed cadence as before.
2. write.ts formatStreamingContent normalized + split('\n') the WHOLE
accumulated content on every 30Hz reveal tick (renderCall also ran a
full-payload normalize first): O(N) per tick, O(N^2) per stream, per
concurrent writer. The collapsed tail-window path now tracks the
newline count incrementally (append-only resume keyed on the
component's persistent render-state object via WeakMap) and extracts
only the tail window with a backward scan: O(delta + preview lines)
per tick, byte-identical output to the split-based reference. The
expanded (Ctrl+O) path is unchanged in output and skips its
split+join round-trip.
Tests: geometric-gate bounds + freshness + small-buffer cadence in
parse-streaming-json-throttled.test.ts; append-growth/reference-
window/CRLF/trailing-newline/restart battery in
write-streaming-incremental.test.ts. Full write/tool-render battery
(56 tests) and ai streaming-args tests (117 tests) pass.
- The two-child fs.watch/poll choreography deadlocked under parallel test load
(runner killed a dangling probe); the PID-namespace pruning test still covers
the multiprocess audit/rotation file contract.
- mnemopi provider parity 'diagnose, validate, graph' does ~6.7s of real
work under bun --parallel=8 on loaded runners; raised its per-test
timeout to 30s (default 5s flaked twice in three CI runs).
- utils LRUCache updateAgeOnGet drove a 30ms TTL with real 20ms sleeps
(10ms margin); now drives performance.now() via a mocked clock, so the
contract is asserted deterministically with no wall-clock wait.
- A blank run at EOF now breaks the list without consuming the blank,
matching real marked: '- item\n\n' lexes as a tight list plus a space
token instead of a loose list whose raw includes the blank.
- Completes the 17.2.10 mid-document fix; same-marker continuation and
indented item content across blanks are unaffected.
- Added list/blank boundary token-shape tests (verified against marked
v15) since the tui incremental tests compare the lexer to itself.
Sunil Srivatsa