mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-25 17:45:53 +01:00
Priya RamanandPaperclip
65a8cecf31
The pre-READY readiness gate searched the full accumulated buffer with
buffer.indexOf("\\n") for every chunk. String#indexOf is linear in the
searched string, so many small newline-less chunks made the cumulative
scan work quadratic in the bytes received before the cap fired. An
untrusted provider could monopolize a host worker CPU.
Record the buffer length before each append and search from that offset
with buffer.indexOf("\\n", from). A newline can only be inside the new
chunk, so the search is correct and the total scan work is linear. Reset
the offset to 0 after the blank-line skip consumes a line.
Add a regression test that counts the scan work through a test-only seam
and asserts it stays proportional to the bytes received. The O(1) buffer
cap, the protocol_contamination result, the settled byte-drop guard, the
nonce check, and the file-bridge fallback stay unchanged.
Co-authored-by: Paperclip <noreply@paperclip.ing>