On 10/9/26 2:55 PM, Collin Walling wrote: > Changelog: > v2 > - updated loop structure to restart and check if previous dependencies > need to be accounted for
To clarify, this was in response to a comment I made to Collin off-list that I failed to mirror here. My concern was that v1 was handling a case where A was deprecated and B depended on A -- but was not catching something like C depending on B which depended on A -- while B would be marked as deprecated, it would not be added to feats[], so C would ultimately not be marked deprecated because B would not be found in feats[]. His v2 patch 2 not only loops to ensure we catch dependencies that may not yet have been marked deprecated on the previous pass, but it also instead looks directly at the most up-to-date features bitmap instead of the local feats[]. AFAICT this is a theoretical problem today, but a potential future hazard since we definitely have such daisy-chained feature dependencies.
