1560049 - Timespan and TimingDistributions are fundamentally bugged: hashCode is not stable outside of obj lifetime

Status: RESOLVED FIXED

(Data Platform and Tools :: Glean: SDK, defect, P1)

Description

[Alessio Placitelli [:Dexter]]

Timespans and TiminigDistirbution are accepting an (id: Any) object to distinguish between running time measurements. However, this relies on hashCode to be stable after the lifetime of 'id', which is not guaranteed.

Luckily for us, no one is using TimingDistribution and Timespans for now (except Glean SDK internals with the 'baseline.duration' metric), so we could fix this with no repercussions in the wild.

The proposed solution:

• have the 'timer.start()' method of both Timespan and TimingDistribution return an handler/integer;
• have the caller manage that timer id and pass it to the 'stopAndSum' when required.

The id would just be a synchronized incremental integer, incremented on every start() call, on the TimingManager.

Comment 2

[Alessio Placitelli [:Dexter]]

Attachment: GitHub Pull Request

Comment 3

[Alessio Placitelli [:Dexter]]

For the future: some of our notes for the reworked API are here - https://docs.google.com/document/d/1x5Vm21K805eN9xyXHt5zvVkR9m-usOyGWQFJoumLE5w/edit