All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=666102

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgy_CKsQsM


Bot(s) for this bug's original alert(s):

win

+grt for win sizes ownership

This increase was due to growth in chrome_child.dll. It appears to have been caused by r432615 ("Switching from CLD3 to CLD2") and then resolved by r432769 ("Switching from CLD2 to CLD3").

I'm marking this as fixed since the size has dropped back down.

abakalov: Was this growth expected? Would you be so kind as to monitor the size of chrome_child.dll if you plan to go back to CLD2? Thanks!

Thanks for looking into this!

Some context -- CLD2 and CLD3 are, respectively, an old and a new language detector. The new one reduces the binary 600K for mobile and 1MB for desktop. We decided for sanity-checking purposes to create and evaluate on an additional dataset. However, it took longer to create that dataset than expected, so we wrote the first CL you mentioned to merge it into M55, so that we switch back to CLD2. The second CL enables the model again, so that we can continue testing it for M56.

If I understand correctly, chrome_child.dll stores data and source code. So, I am wondering:
- What is the regression relative to? M56? If so, then this switch probably caused the regression.
- There is no way of avoiding the increase if we have to make a similar switch involving CLD2. Should we notify someone if this is needed?
- Also, this should have affected all platforms but I just see Windows?

Thanks,
Anton

The measurements here are all relative to the previous continuous build, so it means that the switch back to CLD2 caused a size hit, and going back to CLD3 removed that size hit.

To see the impact on the CL that was merged to M55, we'll need to look at build outputs from that branch. 55.0.2883.54 to 55.0.2883.55:

chrome_child.dll 50231296 bytes to 51394048 bytes 2.31% increase
mini_installer.exe 41050112 bytes to 41758720 bytes 1.73% increase

+kerz in case he has anything to say about whether this is anything to be concerned about, and to chime in on who to notify -- my guess is PMO.

Maybe it did impact other platforms, but it was below the alerting threshold. I shrug.

Adding Alex and Ben for their thoughts.

Adding Simon. IIUC, we don't report APK size in SH Report. Also, comment #3 sounds like this is not a thing we should be talking about as the regression is recovered. Am I missing something?