Incorrect PR Size label

[saad-ali]

In PR kubernetes/kubernetes#33944 the bot correctly applied the size/XXL label to the PR that touches 72 files and adds 6936 lines.

It later removed that label and replaced it with size/XS <-- Buggy behavior

[saad-ali]

Another example where the bot seems to be fighting itself: PR kubernetes/kubernetes#28599

[foxish]

@eparis any idea why this might happen? A github api glitch? I would guess so, because it is adding XS which means it thinks the number of lines changed is < 10.

[eparis]

The sizer is supposed to ignore all changes in vendor so it should be possible for a HUGE PR that only changes vendor to be XS.

But it is obviously broken... Somehow the amount of changes in vendor are being mis-counted...

[eparis]

Seems definitely like a github API bug...

I'm looking specifically at kubernetes/kubernetes#28599.

If I look at the api
https://api.github.com/repos/kubernetes/kubernetes/pulls/28599

I'll see that is says:

  "additions": 18570,
  "deletions": 239,

But if I look at it via git:

git diff --stat $(git merge-base upstream/pr/28599 upstream/master)..upstream/pr/28599

[snip]
 167 files changed, 62513 insertions(+), 41876 deletions(-)

So the github api says there are only 18k insertions but there are really 65k insetions! When we then remove the additions/deletions from ignores paths (in this case that includes at least vendor and more importantly changes to pkg/api/types.generated.go and pkg/api/v1/types.generated.go which BOTH have > 20k additions and deletions) we end up with a negative size.

Additions: -39003 Subtractions: -40594

No idea why the github api just lies about the # of Additions and Deletions.

[eparis]

e-mail sent to github support

[foxish]

@eparis Thanks for figuring it out. I wonder it is a recent change on their end that caused this.

[eparis]

Hi Eric,

If you check https://github.com/kubernetes/kubernetes/pull/28599/files you'll see that the UI shows the same thing (so it's not a problem in the API) and offers an explanation about why the numbers are like that:

"Sorry, we could not display the entire diff because it was too big."

And for that file pkg/api/types.generated.go specifically you'll see:

"20,953 additions, 20,281 deletions not shown because the diff is too large. Please use a local Git client to view these changes."

You're hitting some documented limits in displaying and counting diffs:

https://help.github.com/articles/what-are-the-limits-for-viewing-content-and-diffs-in-my-repository

It's something we'd like to improve in the future, but I can't promise an ETA.

Cheers,
Ivan

[eparis]

Maybe we should calculate the size summing the file.Additions instead of using the pr.Additions and then subtracting out the file.Additions we wish to ignore?