Taylor Blau [off-list ref] writes:

On Mon, Nov 11, 2024 at 12:14:07PM +0100, Karthik Nayak wrote:

quoted

@@ -1697,6 +1701,9 @@ void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
 	struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
 	int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
 	int base_from_cache = 0;
+	unsigned long delta_base_cache_limit = DEFAULT_DELTA_BASE_CACHE_LIMIT;
+
+	repo_config_get_ulong(r, "core.deltabasecachelimit", &delta_base_cache_limit);

 	write_pack_access_log(p, obj_offset);

Hmm. This repo_config_get_ulong() call will look for the configset entry
in a hashmap which is faster than parsing the configuration file from
scratch every time, but still expensive for my taste in a function as
hot as unpack_entry().

Thanks for pointing out, I do have not much idea about the object
database code, so any input here is appreciated.

Should this also go in the_repository->settings instead? That way we
have a single field access instead of a hashmap lookup (with multiple
layers of function calls between us and the actual lookup).

I think that is the best way, though we'll have to still add the config
to the gc config struct. Since there is no repository available there,
but I think it still gets us one step closer towards cleaner config.

Thanks,
Taylor