perf: cut allocations in firehose, archive and Spaces decoding #205

Merged

Grandiras merged 1 commit from issue-195-streaming-perf into main 2026-10-05 12:44:10 +00:00

Conversation 14 Commits 1 Files changed 11 +167 -169

Grandiras commented 2026-10-05 02:47:11 +00:00

Owner

Copy link

Closes #195.

What changed

• Reused transcoder buffer (DagCborJson, DagCborDecoder, EventStreamFrame). One ArrayBufferWriter per thread, taken and returned synchronously around each decode and dropped if it grows past 1 MB. It replaces a fresh buffer of 2× the frame per frame: a 2 MB commit used to allocate 4 MB every time. Reads are bounded by WrittenSpan, and a test covers decoding after a failed decode.
• No string per key or value. Map keys, text values and byte strings are written straight from their UTF-8 bytes. Invalid UTF-8 is replaced exactly as before (new test; a 20,000-case randomized comparison against the old path found 0 differences).
• Frame headers compare keys as bytes and reuse a shared string for the known #commit/#sync/… types.
• Spaces. Record validation reads definite-length byte spans; the repo CAR compares binary CIDs directly.
• RecordProof builds the node CID string only for an error message.
• Debloat. DagCborEncoder gets one $link/$bytes helper. AtProtoCrypto drops its padded big-endian comparer in favour of SequenceCompareTo: every caller checks the 64-byte P1363 length first, and the order constants are fixed 32-byte arrays, so the comparison is numeric. LtHash gets one lane combiner.

Behaviour note

The transcoder now rejects indefinite-length text and byte strings in frames. DAG-CBOR forbids them, and the stricter readers (ReadSeq, TryScan, MstNodeData) already did, so such a frame is dropped as malformed instead of accepted in lax mode.

Measurements (BenchmarkDotNet, in-process, baseline origin/main)

Scenario	Baseline	Candidate
Firehose consume, 100 commit frames	515 µs, 1839 KB	420 µs, 675 KB
Same, with Sync 1.1 verification	25.1 ms, 4725 KB	25.1 ms, 3561 KB
DAG-CBOR decode of a post ×100	160 µs, 316 KB	140 µs, 179 KB
Spaces record validation ×100	65 µs, 170 KB	56 µs, 87 KB

Verification time is dominated by native ECDSA (~206 of ~240 µs per commit).

Net src: −25 lines. Build has 0 warnings; 4966 tests pass.

🤖 Generated with Claude Code

Closes #195. ## What changed - **Reused transcoder buffer** (`DagCborJson`, `DagCborDecoder`, `EventStreamFrame`). One `ArrayBufferWriter` per thread, taken and returned synchronously around each decode and dropped if it grows past 1 MB. It replaces a fresh buffer of 2× the frame per frame: a 2 MB commit used to allocate 4 MB every time. Reads are bounded by `WrittenSpan`, and a test covers decoding after a failed decode. - **No string per key or value.** Map keys, text values and byte strings are written straight from their UTF-8 bytes. Invalid UTF-8 is replaced exactly as before (new test; a 20,000-case randomized comparison against the old path found 0 differences). - **Frame headers** compare keys as bytes and reuse a shared string for the known `#commit`/`#sync`/… types. - **Spaces.** Record validation reads definite-length byte spans; the repo CAR compares binary CIDs directly. - **`RecordProof`** builds the node CID string only for an error message. - **Debloat.** `DagCborEncoder` gets one `$link`/`$bytes` helper. `AtProtoCrypto` drops its padded big-endian comparer in favour of `SequenceCompareTo`: every caller checks the 64-byte P1363 length first, and the order constants are fixed 32-byte arrays, so the comparison is numeric. `LtHash` gets one lane combiner. ## Behaviour note The transcoder now rejects indefinite-length text and byte strings in frames. DAG-CBOR forbids them, and the stricter readers (`ReadSeq`, `TryScan`, `MstNodeData`) already did, so such a frame is dropped as malformed instead of accepted in lax mode. ## Measurements (BenchmarkDotNet, in-process, baseline `origin/main`) | Scenario | Baseline | Candidate | |---|---|---| | Firehose consume, 100 commit frames | 515 µs, 1839 KB | 420 µs, 675 KB | | Same, with Sync 1.1 verification | 25.1 ms, 4725 KB | 25.1 ms, 3561 KB | | DAG-CBOR decode of a post ×100 | 160 µs, 316 KB | 140 µs, 179 KB | | Spaces record validation ×100 | 65 µs, 170 KB | 56 µs, 87 KB | Verification time is dominated by native ECDSA (~206 of ~240 µs per commit). Net `src`: −25 lines. Build has 0 warnings; 4966 tests pass. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Grandiras added 1 commit 2026-10-05 02:47:11 +00:00

perf: cut allocations in firehose, archive and Spaces decoding (closes #195) ... 32fb60b1d2

The DAG-CBOR to JSON transcoder reuses a per-thread buffer and writes keys,
text and byte strings from their UTF-8 instead of through a string or array
each, so a #commit frame allocates about a third of what it did. The frame
header shares its type strings, Spaces record validation reads without
copying, and the Spaces repo CAR no longer round-trips CIDs through strings.
Small debloat: one lane combiner in LtHash, no padded big-endian compare in
the signature range checks, and a shared helper for the $link/$bytes forms
in the DAG-CBOR encoder.

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

opencode-bot commented 2026-10-05 02:48:21 +00:00

Collaborator

Copy link

🤖 Reviewing this PR — the review will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the review runs.

🤖 Reviewing this PR — the review will link the full session transcript. <!-- opencode-run:review --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T02-48-21-629Z.md) — updating every ~5 minutes while the review runs.

Grandiras referenced this pull request 2026-10-05 02:53:13 +00:00

refactor: debloat the Spaces server, EF Core stores and streaming consumers #207

opencode-bot commented 2026-10-05 02:57:12 +00:00

Collaborator

Copy link

🤖 Resolving the merge conflicts — the reply will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the fix runs.

🤖 Resolving the merge conflicts — the reply will link the full session transcript. <!-- opencode-run:conflict --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T02-57-11-766Z.md) — updating every ~5 minutes while the fix runs.

Grandiras commented 2026-10-05 02:57:25 +00:00

Author

Owner

Copy link

Independent adversarial review (read-only Sonnet reviewer): no defects found. A differential probe of DagCborDecoder.Decode against main gave byte-identical output on 9 cases: invalid UTF-8 variants, indefinite lengths, escapes, duplicate keys and key order. The reused buffer is synchronous-only, nested-safe and returned in finally, and nothing references it after return. Every SequenceCompareTo caller is gated on the 64-byte length, and the order constants are 32 bytes. LtHash.Combine matches the old add/subtract lanes. Unit suite: 4966 passed.

Independent adversarial review (read-only Sonnet reviewer): **no defects found**. A differential probe of `DagCborDecoder.Decode` against `main` gave byte-identical output on 9 cases: invalid UTF-8 variants, indefinite lengths, escapes, duplicate keys and key order. The reused buffer is synchronous-only, nested-safe and returned in `finally`, and nothing references it after return. Every `SequenceCompareTo` caller is gated on the 64-byte length, and the order constants are 32 bytes. `LtHash.Combine` matches the old add/subtract lanes. Unit suite: 4966 passed.

Grandiras force-pushed issue-195-streaming-perf from 32fb60b1d2 to 33ae74e1a6 2026-10-05 02:58:18 +00:00 Compare

opencode-bot commented 2026-10-05 02:58:19 +00:00

Collaborator

Copy link

⚠️ The conflict run failed: git -c http.extraheader=Authorization: Basic [REDACTED] push origin HEAD:refs/heads/issue-195-streaming-perf failed (1): To http://forgejo:3000/Grandiras/ATProto.NET.git
! [rejected] HEAD -> issue-195-streaming-perf (fetch first)
error: failed to push some refs to 'http://forgejo:3000/Grandiras/ATProto.NET.git'
hint: Updates were rejected because the remote contains work that you do
hint: not have locally. T

⚠️ The conflict run failed: git -c http.extraheader=Authorization: Basic [REDACTED] push origin HEAD:refs/heads/issue-195-streaming-perf failed (1): To http://forgejo:3000/Grandiras/ATProto.NET.git ! [rejected] HEAD -> issue-195-streaming-perf (fetch first) error: failed to push some refs to 'http://forgejo:3000/Grandiras/ATProto.NET.git' hint: Updates were rejected because the remote contains work that you do hint: not have locally. T

Grandiras referenced this pull request 2026-10-05 03:04:53 +00:00

refactor: debloat the Spaces server, EF Core stores and streaming consumers #208

Grandiras commented 2026-10-05 11:49:30 +00:00

Author

Owner

Copy link

@claude /review

@claude /review

claude-bot commented 2026-10-05 11:50:01 +00:00

Collaborator

Copy link

🤖 Reviewing this PR — follow along in the action run; the reply will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the task runs.

🤖 Reviewing this PR — follow along in [the action run](https://git.grandiras.net/Grandiras/claude-bot/actions/runs/732); the reply will link the full session transcript. <!-- claude-run:pr_review --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-732.md) — updating every ~5 minutes while the task runs.

claude-reviewer approved these changes 2026-10-05 11:50:26 +00:00

Dismissed

claude-reviewer left a comment

Copy link

✅ Verdict: approve — no blocking issues found.

I found no blocking problems. I read the diff and the surrounding code; I did not build the PR or run the tests.

• Thread-local buffer: Rent/return is safe. Every decode is synchronous and the output is consumed before the finally runs. The Utf8JsonWriter is disposed before the buffer is reset. A failed decode leaves nothing behind because ResetWrittenCount runs in finally, and there is a test for that. A nested call gets a fresh buffer because the field is nulled on rent. The 1 MB cap bounds memory held per thread.
• Definite-length reads: ReadDefiniteLengthTextStringBytes and ReadDefiniteLengthByteString raise InvalidOperationException or CborContentException on indefinite-length strings. Every call site (DagCborDecoder, TryReadHeader, Deserialize) already catches those through IsMalformed. The stricter handling is documented in the PR description.
• ReadUtf8: The span stays valid for the writer call. Invalid UTF-8 goes through a string round trip, which reproduces the old replacement behaviour, and a test covers it.
• Crypto: SequenceCompareTo is numerically correct for equal-length big-endian inputs, and NormalizeS now returns early on the same conditions as before.
• LtHash.Combine: Behaviour is equivalent to the two removed helpers.

Minor: a short comment in AtProtoCrypto stating that the SequenceCompareTo calls rely on the callers' 64-byte length check would stop a future caller from quietly breaking the numeric ordering.

---

🧠 claude-sonnet-5-5 · ⏱ 0m 16s · 3 turns · tokens: 27.2k in / 1.8k out (+49.4k cache-read)
📜 Full session transcript

✅ **Verdict: approve** — no blocking issues found. I found no blocking problems. I read the diff and the surrounding code; I did not build the PR or run the tests. - **Thread-local buffer:** Rent/return is safe. Every decode is synchronous and the output is consumed before the `finally` runs. The `Utf8JsonWriter` is disposed before the buffer is reset. A failed decode leaves nothing behind because `ResetWrittenCount` runs in `finally`, and there is a test for that. A nested call gets a fresh buffer because the field is nulled on rent. The 1 MB cap bounds memory held per thread. - **Definite-length reads:** `ReadDefiniteLengthTextStringBytes` and `ReadDefiniteLengthByteString` raise `InvalidOperationException` or `CborContentException` on indefinite-length strings. Every call site (`DagCborDecoder`, `TryReadHeader`, `Deserialize`) already catches those through `IsMalformed`. The stricter handling is documented in the PR description. - **`ReadUtf8`:** The span stays valid for the writer call. Invalid UTF-8 goes through a string round trip, which reproduces the old replacement behaviour, and a test covers it. - **Crypto:** `SequenceCompareTo` is numerically correct for equal-length big-endian inputs, and `NormalizeS` now returns early on the same conditions as before. - **`LtHash.Combine`:** Behaviour is equivalent to the two removed helpers. Minor: a short comment in `AtProtoCrypto` stating that the `SequenceCompareTo` calls rely on the callers' 64-byte length check would stop a future caller from quietly breaking the numeric ordering. --- 🧠 claude-sonnet-5-5 · ⏱ 0m 16s · 3 turns · tokens: 27.2k in / 1.8k out (+49.4k cache-read) 📜 [Full session transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-732.md) <!-- claude-verdict:approve --> <!-- claude-reviewed-head:33ae74e1a623a63ba43d8525cfbee98f3685cb86 -->

Grandiras force-pushed issue-195-streaming-perf from 33ae74e1a6 to 9533157f8c 2026-10-05 12:31:23 +00:00 Compare

opencode-bot commented 2026-10-05 12:32:25 +00:00

Collaborator

Copy link

🤖 Reviewing this PR — the review will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the review runs.

🤖 Reviewing this PR — the review will link the full session transcript. <!-- opencode-run:review --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T12-32-25-007Z.md) — updating every ~5 minutes while the review runs.

claude-bot commented 2026-10-05 12:32:29 +00:00

Collaborator

Copy link

🤖 Reviewing this PR — follow along in the action run; the reply will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the task runs.

🤖 Reviewing this PR — follow along in [the action run](https://git.grandiras.net/Grandiras/claude-bot/actions/runs/735); the reply will link the full session transcript. <!-- claude-run:pr_review --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-735.md) — updating every ~5 minutes while the task runs.

claude-reviewer approved these changes 2026-10-05 12:32:46 +00:00

claude-reviewer left a comment

Copy link

✅ Verdict: approve — no blocking issues found.

I found no blocking problems. I read the diff and did not build the PR or run the tests.

My earlier pass was an approval with one minor, non-blocking note: a comment in AtProtoCrypto saying that the SequenceCompareTo calls depend on the callers' 64-byte P1363 length check. The diff doesn't add it. It would be a cheap guard against a future caller breaking the numeric ordering, but it doesn't block merging.

Re-checked and still fine:

• Buffer reuse: the thread-local buffer is taken on rent and reset in finally. The writer is disposed before the reset, and a nested call gets a fresh buffer. The test for decoding after a failed decode covers the reset.
• Strict decoding: the ReadDefinite* reads reject indefinite-length strings, and every call site already catches that through its malformed-input handling. The PR description documents the stricter behaviour.
• Invalid UTF-8: the replacement path goes through a string round trip, which keeps the old behaviour, and a new test covers it.
• LtHash.Combine and NormalizeS: both are equivalent to the code they replace.

---

🧠 claude-sonnet-5-5 · ⏱ 0m 6s · 2 turns · tokens: 26.0k in / 759 out (+12.2k cache-read)
📜 Full session transcript

✅ **Verdict: approve** — no blocking issues found. I found no blocking problems. I read the diff and did not build the PR or run the tests. My earlier pass was an approval with one minor, non-blocking note: a comment in `AtProtoCrypto` saying that the `SequenceCompareTo` calls depend on the callers' 64-byte P1363 length check. The diff doesn't add it. It would be a cheap guard against a future caller breaking the numeric ordering, but it doesn't block merging. Re-checked and still fine: - **Buffer reuse:** the thread-local buffer is taken on rent and reset in `finally`. The writer is disposed before the reset, and a nested call gets a fresh buffer. The test for decoding after a failed decode covers the reset. - **Strict decoding:** the `ReadDefinite*` reads reject indefinite-length strings, and every call site already catches that through its malformed-input handling. The PR description documents the stricter behaviour. - **Invalid UTF-8:** the replacement path goes through a string round trip, which keeps the old behaviour, and a new test covers it. - **`LtHash.Combine` and `NormalizeS`:** both are equivalent to the code they replace. --- 🧠 claude-sonnet-5-5 · ⏱ 0m 6s · 2 turns · tokens: 26.0k in / 759 out (+12.2k cache-read) 📜 [Full session transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-735.md) <!-- claude-verdict:approve --> <!-- claude-reviewed-head:9533157f8c880e6111110c976f67a91b49f98aff -->

opencode-reviewer reviewed 2026-10-05 12:39:07 +00:00

opencode-reviewer left a comment

Copy link

💬 Verdict: comment — the reviewer did not submit a structured review; the raw output follows.

Let me check the crypto context for the SequenceCompareTo change, since length mismatches matter there.

---

⏱ 6m 42s · 3 steps · $0.0057 · tokens: 21.0k in / 126 out (+9.0k cache-read) · Go: 5h 0% · wk 0% · mo 91%
📜 Full session transcript

💬 **Verdict: comment** — the reviewer did not submit a structured review; the raw output follows. Let me check the crypto context for the `SequenceCompareTo` change, since length mismatches matter there. --- ⏱ 6m 42s · 3 steps · $0.0057 · tokens: 21.0k in / 126 out (+9.0k cache-read) · Go: 5h 0% · wk 0% · mo 91% 📜 [Full session transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T12-32-25-007Z.md)

Grandiras commented 2026-10-05 12:39:48 +00:00

Author

Owner

Copy link

@opencode-bot /review

@opencode-bot /review

opencode-bot commented 2026-10-05 12:39:49 +00:00

Collaborator

Copy link

🤖 Reviewing this PR — the review will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the review runs.

🤖 Reviewing this PR — the review will link the full session transcript. <!-- opencode-run:review --> 📜 [Live transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T12-39-49-095Z.md) — updating every ~5 minutes while the review runs.

opencode-reviewer approved these changes 2026-10-05 12:40:42 +00:00

opencode-reviewer left a comment

Copy link

✅ Verdict: approve — no blocking issues found.

No blocking problems found. This is a fresh pass (no prior opencode-reviewer findings to resolve); the earlier approvals from claude-reviewer stand under my own checks.

Verified points:

• Thread-local buffer reuse (DagCborJson.RentBuffer/ReturnBuffer): rent/return is synchronous around each decode, the Utf8JsonWriter is disposed before ResetWrittenCount runs in finally, and the output (WrittenSpan → JsonElement.ParseValue / JsonSerializer.Deserialize) is consumed inside the try, so nothing references the reused buffer afterwards. A nested call rents a fresh buffer because the field is nulled on rent. The 1 MB cap bounds what a thread retains, and the new Decode_AfterAFailedDecode_ReturnsOnlyItsOwnValue test covers a failed decode leaving no residue.
• SequenceCompareTo in AtProtoCrypto: every reachable caller checks the 64-byte P1363 signature length first — AtProtoKey.Verify (via HasSignatureLength), DidKeyCache.VerifyJws, and AtProtoKey.Sign via the fixed-field signature format — and OrderBytes/HalfOrder are fixed 32-byte big-endian arrays, so byte-wise lexicographic order equals numeric order. NormalizeLowSSignature's restructure preserves the old behaviour for the reachable inputs. The minor suggestion to add a comment stating the 64-byte length precondition (raised in previous passes) is still worth a line, but it remains non-blocking.
• Strict definite-length reads (ReadDefiniteLengthTextStringBytes/ReadDefiniteLengthByteString): indefinite-length strings now throw, and every call site (DagCborDecoder.Decode, EventStreamFrame.TryReadHeader/Deserialize, SpaceRecordCbor) already filters malformed input through IsMalformed or its equivalent, so an indefinite-length frame is dropped as malformed rather than crashing. The behaviour change is documented in the PR description.
• ReadUtf8 invalid-UTF-8 fallback: goes through a string round trip so replacement matches what the lax CborReader produced before, and the returned span is written immediately. The randomized comparison plus the new test give good coverage.
• LtHash.Combine, RecordProof.Name (error-path-only string), SpaceRepoCar binary CID comparison, and the DagCborEncoder.TryGetOnlyString helper: all behaviour-equivalent to what they replace.

---

⏱ 0m 53s · 10 steps · $0.0129 · tokens: 30.6k in / 1.1k out (+140.2k cache-read) · Go: 5h 0% · wk 0% · mo 91%
📜 Full session transcript

✅ **Verdict: approve** — no blocking issues found. No blocking problems found. This is a fresh pass (no prior opencode-reviewer findings to resolve); the earlier approvals from claude-reviewer stand under my own checks. Verified points: - **Thread-local buffer reuse** (`DagCborJson.RentBuffer`/`ReturnBuffer`): rent/return is synchronous around each decode, the `Utf8JsonWriter` is disposed before `ResetWrittenCount` runs in `finally`, and the output (`WrittenSpan` → `JsonElement.ParseValue` / `JsonSerializer.Deserialize`) is consumed inside the `try`, so nothing references the reused buffer afterwards. A nested call rents a fresh buffer because the field is nulled on rent. The 1 MB cap bounds what a thread retains, and the new `Decode_AfterAFailedDecode_ReturnsOnlyItsOwnValue` test covers a failed decode leaving no residue. - **`SequenceCompareTo` in `AtProtoCrypto`**: every reachable caller checks the 64-byte P1363 signature length first — `AtProtoKey.Verify` (via `HasSignatureLength`), `DidKeyCache.VerifyJws`, and `AtProtoKey.Sign` via the fixed-field signature format — and `OrderBytes`/`HalfOrder` are fixed 32-byte big-endian arrays, so byte-wise lexicographic order equals numeric order. `NormalizeLowSSignature`'s restructure preserves the old behaviour for the reachable inputs. The minor suggestion to add a comment stating the 64-byte length precondition (raised in previous passes) is still worth a line, but it remains non-blocking. - **Strict definite-length reads** (`ReadDefiniteLengthTextStringBytes`/`ReadDefiniteLengthByteString`): indefinite-length strings now throw, and every call site (`DagCborDecoder.Decode`, `EventStreamFrame.TryReadHeader`/`Deserialize`, `SpaceRecordCbor`) already filters malformed input through `IsMalformed` or its equivalent, so an indefinite-length frame is dropped as malformed rather than crashing. The behaviour change is documented in the PR description. - **`ReadUtf8` invalid-UTF-8 fallback**: goes through a string round trip so replacement matches what the lax `CborReader` produced before, and the returned span is written immediately. The randomized comparison plus the new test give good coverage. - **`LtHash.Combine`**, **`RecordProof.Name`** (error-path-only string), **`SpaceRepoCar`** binary CID comparison, and the **`DagCborEncoder.TryGetOnlyString`** helper: all behaviour-equivalent to what they replace. --- ⏱ 0m 53s · 10 steps · $0.0129 · tokens: 30.6k in / 1.1k out (+140.2k cache-read) · Go: 5h 0% · wk 0% · mo 91% 📜 [Full session transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-205/run-2026-10-05T12-39-49-095Z.md) <!-- opencode-reviewed-head:9533157f8c880e6111110c976f67a91b49f98aff --> <!-- opencode-verdict:approve -->

Grandiras merged commit 0b0a4a9c13 into main 2026-10-05 12:44:10 +00:00

Grandiras deleted branch issue-195-streaming-perf 2026-10-05 12:44:10 +00:00

Grandiras referenced this pull request from a commit 2026-10-05 12:44:11 +00:00

perf: cut allocations in firehose, archive and Spaces decoding (#205)

Sign in to join this conversation.

Reviewers

No reviewers

claude-reviewer

opencode-reviewer

Labels

Clear labels   

breaking-change

Introduces a breaking API change

  

bug

Something is not working correctly

  

documentation

Documentation improvements

  

duplicate

This issue or PR already exists

  

enhancement

New feature or improvement

  

good first issue

Good for newcomers to the project

  

help wanted

Extra attention is needed

  

performance

Performance-related issue or improvement

  

question

Further information is requested

  

wontfix

This will not be worked on

No labels

breaking-change

bug

documentation

duplicate

enhancement

good first issue

help wanted

performance

question

wontfix

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

5 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

Grandiras/ATProto.NET!205

No description provided.