sscoble/pi-mono
1
0
Fork 0
mirror of https://github.com/badlogic/pi-mono.git synced 2026-11-05 23:37:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

docs(agent): record JSONL fork capture question

Browse source 
Contrast the original fixed-length source capture requirement with the prototype's next-sequence boundary and document the file-replacement tradeoff.
This commit is contained in:
David Brailovsky 2026-09-04 16:11:41 +02:00
parent 9ee703230f
commit 420deedf81
1 changed files with 8 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

8
packages/agent/docs/work-packages/08-named-branch-streaming-forks.md
View file

@ -250,3 +250,11 @@ Current: reject any malformed lane. Possible relaxation: trust supported atomic
### 9.2 Sequence preservation
Current: preserve source sequences, `nextSeq`, and list cursors. Possible relaxation: keep entry sequences but assign destination-local sequences to values, list elements, and reconstructed lane rows. Revisit if preservation materially complicates streaming.
### 9.3 JSONL source capture boundary
Original requirement: capture the exact source file length at the commit-queue boundary and have both passes read only up to that length through the same fixed file handle.
Current implementation: capture the source's current `nextSeq` at the commit-queue boundary and have both independently opened scans stop before that sequence. This is simpler and avoids adding byte-range semantics to `FileSystem`, while still excluding later append-only commits transaction by transaction.
The current implementation assumes a format-4 source remains append-only and is not replaced while the fork runs. Concurrent compaction or another whole-file replacement could otherwise make the two scans observe different file incarnations. Decide whether to restore fixed-length capture, retain the sequence boundary with an explicit no-replacement invariant, or add replacement coordination when J1 compaction is implemented.