This message was deleted.

We're working on fixing the error messaging here - in most (if not all) failure cases the diff is too big

What are the thresholds for too big? And is it just diff size or is the size of the file modified also a factor (irrespective of diff size)?

only diff size: unchanged lines in a file aren't counted. as for threshold, right now I'd say diffs that exceed a very rough ballpark of 100 lines are more likely to hit the cap, especially if the lines themselves are long I plan to look more closely at failure cases to identify a clearer line to communicate in the app. the root cap is from openai token limits, which is proportional but not identical to character count. we compress the diff itself by stripping out or simplifying lines that we expect to not impact the summarization quality, so there's high variance in indicating a cap by just line or character count alone.

copy tyvm! 100 lines is pretty tiny! but good to know so i know i can try it on something small