originally the reason for pre-sizing was that that's easier for a multi-threaded use of the hash table. That hasn't materialized yet, so there's not much sense in using the very very conservative estimates for pre-sizing. Doing the resize on-demand, whenever we actually need to add a new entry doesn't change performance. bfd/ merge.c (sec_merge_hash_insert): Resize as needed from here ... (record_section): ... not from here. Don't calculate estimates, return bool instead of three-state, regard all errors as soft errors. (_bfd_merge_sections): Adjust.