On 11/7/2019 12:32 AM, Junio C Hamano wrote:
> "Erik Chen via GitGitGadget" <gitgitgadget@xxxxxxxxx> writes:
>
>> From: Erik Chen <erikchen@xxxxxxxxxxxx>
>>
>> Add trace2 regions to fetch-pack.c to better track time spent in the various
>> phases of a fetch:
>>
>> * matching common remote and local refs
>> * marking local refs as complete (part of the matching process)
>>
>> Both of these stages can be slow for repositories with many refs.
>>
>> Signed-off-by: Erik Chen <erikchen@xxxxxxxxxxxx>
>> ---
>> fetch-pack.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>
> OK.
>
>> diff --git a/fetch-pack.c b/fetch-pack.c
>> index 0130b44112..5e3eee0477 100644
>> --- a/fetch-pack.c
>> +++ b/fetch-pack.c
>> @@ -669,6 +669,8 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator,
>>
>> save_commit_buffer = 0;
>>
>> + trace2_region_enter("fetch-pack", "mark_complete_and_common_ref", NULL);
>> +
>> for (ref = *refs; ref; ref = ref->next) {
>> struct object *o;
>>
>> @@ -690,6 +692,10 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator,
>> }
>> }
>>
>> + /* This block marks all local refs as COMPLETE, and then recursively marks all
>> + * parents of those refs as COMPLETE.
>> + */
>
> /*
> * We write our multi-line comments like this, with the
> * slash-asterisk at the beginning and the asterisk-slash
> * at the end on its own line. Learn such local conventions
> * from the existing surrounding code and imitate, which
> * would reduce stylistic errors.
> */
>
> Will fix-up while queuing (no need to reroll only to fix this).
>
>> + trace2_region_enter("fetch-pack", "mark_complete_local_refs", NULL);
>> if (!args->deepen) {
>> for_each_ref(mark_complete_oid, NULL);
>> for_each_cached_alternate(NULL, mark_alternate_complete);
>> @@ -697,6 +703,7 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator,
>> if (cutoff)
>> mark_recent_complete_commits(args, cutoff);
>> }
>> + trace2_region_leave("fetch-pack", "mark_complete_local_refs", NULL);
>>
>> /*
>> * Mark all complete remote refs as common refs.
>> @@ -716,6 +723,7 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator,
>> }
>>
>> save_commit_buffer = old_save_commit_buffer;
>> + trace2_region_leave("fetch-pack", "mark_complete_and_common_ref", NULL);
>
>
> So this introduces a single region around the entire function body
> of mark_complete_and_common_ref(), within which only one subpart is
> also enclosed in a nested region. Is that because the parts inside
> the outer region before and after the inner region are known to
> consume negligible time? IOW I would understand
>
> F () {
> <region 1 begin>
> <region 1.1 begin>
> code
> <region 1.1 end>
> <region 1.2 begin>
> code
> <region 1.2 end>
> <region 1.3 begin>
> code
> <region 1.3 end>
> <region 1 end>
> }
>
> or
>
> F () {
> trivial code
> <region 1 begin>
> heavy code
> <region 1 end>
> trivial code
> }
>
> but this appears to do
>
>
> F () {
> <region 1 begin>
> code
> <region 1.1 begin>
> code
> <region 1.1 end>
> code
> <region 1 end>
> }
>
> which is somewhat puzzling.
I notice that a v4 was sent that adds more sub-regions without actually
responding to this request. (It is worth also pointing out that you
ignored Junio's request you use the cover letter to explain your reasoning
for changes between versions.)
There is a real downside to nesting regions like this. Specifically, we
frequently limit the depth that we report nested regions to avoid
overwhelming the logs.
In general, these sub-regions should be avoided when possible and instead
create regions around important sections, such as the second option Junio
lists above.
Thanks,
-Stolee
