mirror of
https://github.com/nearai/ironclaw.git
synced 2026-09-02 23:56:24 +08:00
0a8428fda1
* fix(telegram): handle "message is too long" with retry splitting Reduce TELEGRAM_MAX_MESSAGE_LEN from 4096 to 4000 for safety margin against Markdown entity/emoji counting edge cases. Add SendError::TooLong variant and send_chunk() helper that recursively halves chunks on "message is too long" rejections (up to 3 levels deep), splitting at natural boundaries. [skip-regression-check] * fix: address review feedback — return last chunk id, handle TooLong on plain-text retry - Extract split_and_send() helper returning last message_id for correct reply threading when chunks are split - Handle TooLong on ParseEntities plain-text fallback path - Update doc comments on send_message and split_message [skip-regression-check] * fix: address review feedback — UTF-16 split, depth cap, markdown flag, tests - Use UTF-16 code units (via prefix_within_utf16_limit) for midpoint calculation in split_and_send, matching Telegram's actual limit - Extract find_split_midpoint() as a pure testable function - Reduce MAX_SPLIT_DEPTH from 3 to 2 (max 4 sub-messages per chunk) - Pass use_markdown flag through recursion so ParseEntities fallback disables Markdown for all subsequent splits of that chunk - Add debug logging on successful sends - Improve error message when depth limit exhausted - Add empty-text guard in split_and_send - Add 5 unit tests for find_split_midpoint (paragraph, newline, space, no-boundary, emoji-heavy) - Fix stale doc comments * fix: guard against empty first half after trim in split_and_send Whitespace-heavy text could produce an empty first half after trim_end(). Skip directly to the second half in that case. * refactor(telegram): unify message splitting into single parameterized splitter Collapse the duplicate boundary-search logic in find_split_midpoint / split_and_send into split_message by parameterizing its UTF-16 limit. The TooLong retry path now calls split_message(text, limit/2) and sends each sub-chunk, so the retry benefits from the same paragraph → newline → sentence → word hierarchy the initial split already used. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Illia Polosukhin <ilblackdragon@gmail.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>