Git 2.55 introduces incremental repacking for high-volume repositories

A repository of sufficient scale eventually encounters a physics problem: the cost of maintaining its own metadata. When a system indexes every data pack in a single file, even a minor update requires a total rewrite of that index. Git 2.55 addresses this by formalizing incremental multi-pack indexes, allowing the system to store its lookup data as a chain of layers. New data is indexed at the tip of the chain, leaving the deeper, larger layers undisturbed.

This release enables the direct writing of these incremental chains during routine repacking. To prevent the chain itself from becoming a burden, the system now supports geometric repacking rules that decide when adjacent layers should be compacted based on their relative size. When a new layer grows large enough relative to its predecessor, Git merges their metadata into a single replacement layer. This compaction remains metadata-only; it does not require the expensive re-processing of underlying object files, effectively decoupling the speed of repository maintenance from its total size.