On 07/01/20(Tue) 18:57, Stefan Sperling wrote:
> On Tue, Jan 07, 2020 at 01:02:13PM +0100, Martin Pieuchot wrote:
> > got_packidx_init_hdr() is also taking a lot of CPU time.  Is it because
> > of the SHA operations?  Is git using the sames?  Can they be improved?
> 
> The answer is that the pack index caching code is broken.
> This function was called way too often: 749 times for 14 pack files.
> The diff below fixes this bug and reduces this function to 2.2%.
> Updated profile graph attached.
> However, this patch only has a very small impact on performance.

Maybe.  On the other hand the profiled picture becomes clearer: now ~36%
of CPU time is spent diffing...

One question below:

> @@ -129,6 +129,7 @@
>  #define GOT_ERR_REF_NAME_MINUS	113
>  #define GOT_ERR_GITCONFIG_SYNTAX 114
>  #define GOT_ERR_REBASE_OUT_OF_DATE 115
> +#define GOT_ERR_CACHE_DUP_ENTRY	116
>  
>  static const struct got_error {
>  	int code;
> @@ -265,6 +266,7 @@ static const struct got_error {
>  	{ GOT_ERR_GITCONFIG_SYNTAX, "gitconfig syntax error" },
>  	{ GOT_ERR_REBASE_OUT_OF_DATE, "work tree must be updated before it "
>  	    "can be used to rebase a branch" },
> +	{ GOT_ERR_CACHE_DUP_ENTRY, "duplicate cache entry" },

Isn't a duplicated cache entry some kind of code error?  Or this can
happen in a repo?