Web lists-archives.com

Re: [RFC PATCH 000/194] Moving global state into the repository object




On Tue, Feb 6, 2018 at 3:25 PM, Stefan Beller <sbeller@xxxxxxxxxx> wrote:
>> Any suggestions welcome!
>
> Eric repeatedly points out leaking memory.
>
> As of today we do not care about memory leaking as it is cleaned
> up at the end of the program anyway, for example the objects
> hash table is never cleared.

Is this still true/desirable when multiple 'repos' are involved?

> In a resend I will put the infrastructure in place to free the memory via
> adding
>
>   raw_object_store_clear(struct raw_object_store *)
>   object_parser_clear(struct object_parser *)
>   ref_store_clear(struct ref_store *)
>
> and calling these functions on repo destruction. The functions
> itself will be empty code-wise and contain TODO comments listing
> all variables that need care.

I'm confused. If you go to the effort of inserting TODO's why not go
all the way and instead insert the actual code?

> Follow up patches can figure out what is best to do, such as closing
> the memleaks. As repo_clear is not called for the_repository
> we'd even keep the property of relying on fast cleanup by the OS.