Re: Idea: "Enable" function for stats members?

Den 09. jan. 2017 01:33, skrev Martin Thomson:
> On 9 January 2017 at 09:24, Harald Alvestrand <> wrote:
>> Den 08. jan. 2017 23:19, skrev Martin Thomson:
>>> On 7 January 2017 at 07:15, Jan-Ivar Bruaroey <> wrote:
>>>> This might be a bigger change to the spec than implementations, as it might
>>>> mean converting the dictionaries to interfaces...
>>> This is something I'd like to see happen.
>> If the interface remains iterable, people who collect "all the stats"
>> will still hit all the items, making usage counts useless.
> Those uses would create a floor on the measurements, so you just have
> to correct for that.  (I suppose it might be possible to count
> accesses to the iterator to help with that correction.)

We could use the mechanisms we usually do for experimental APIs to
release new stats into the world, too. Hm.

Do people who collect stats use a serialization (such as JSON) when they
collect stats? That should Just Work, if we put "serializer=attributes".

I'm not convinced of the advantages, but it doesn't seem impossible.

Received on Monday, 9 January 2017 22:13:19 UTC