The workaround is that if the scale-factor is uniform along all axes (x, y, z), then the collision object seem to scale correctly. I encountered the same issue with scaling sprites and connected trigger object.
This is mentioned by @sicher a bit down the thread, but not explicitly mentioned as a workaround for the issue, so…
Am I missing something?
I also bumped in this problem. Luckily, I could use uniform scaling.
I also noticed, that my particle.fx got scaled once i did uniform scaling.
FYI, I just bumped into this too.
What caught me was that the collision box that I was using appeared scaled in the editor pane, but then when I turned on physics debugging it wasn’t.
The exact same thing as described by @rantingbob happened to me today. Scaling the parent affects the box in the editor, but not in the game (as seen by turning on physics debug).
@britzl said it was moved to bugs in Jira two years ago. Is it still not done? Or has that one been fixed and this should be filed as a new one?
Although it took us a long time before we picked it up, the DEF-521 is actually in progress. It just needs some other tasks first, for us in order to make this a more solid solution. And I don’t have a definitive ETA yet. Sorry for the delay. 
Is this not fixed yet? I just encountered it now.
Parent go has multiple factory-created children go’s, each with its own collision. When scaling the parent go, the children go’s get scaled properly, but the attached collision sizes remain the same.
Sadly, no. Those other (required) tasks haven’t been prioritized during the year.
But I feel hopeful we’ll pick those up at the beginning of next year. (But again, cannot promise an ETA yet)
Having used Box2D a bit more directly through Löve2D, I have a better appreciation for why this “issue” exists — It’s simply the way Box2D works. It is not a bug, in Defold or anything else. Box2D does not allow you to change existing shapes, only to create and destroy them. I assume this is a performance optimization, so any “fix” will cause at least some performance hit.
The real bug I see is that the editor does not respect this. So your game objects can look one way in the editor but completely different in the actual, running game. There have been a few reports for this: here, here, and here.
Just started playing around with Defold and faced this issue, when I scale a GameObject with attached CollisionObject, in the Editor both sprite and CO look scaled correctly, but when I run the game then CO appears scaled incorrectly. I can rescale collision object manually so it will fit GO size, but again, now in the Editor CO looks messed, but at least it’s ok in the game.
Please, make it at least consistent, broken in both places, or fixed in both. Thanks!
The original ticket (DEF-521) is about scaling the game object at runtime.
Your issue seems different. Afaik, the scaling of game objects at edit time will be represented ok in the engine.
Can you give us a repro case to test with? E.g. zip up a small test project and share it here. (Exclude the build folder and any hidden folders)
I had this same problem last week. Here’s a short video and test project to demonstrate the issue (the checkerboard pattern is scaled x 4):
co-scale-test.zip (3.3 KB)
I’m sure box2d doesn’t allow any shape scaling on runtime.But you or the game engine could have some workaround here.
We’re bumping this into our next sprint (in one week). We’ve worked on this on and off in the past and have code that is working, but it’s not a perfect solution that we want to release yet, and as you say box2d doesn’t allow this but we can solve it in the engine. We now feel that we need to deal with this old feature request once and for all. We’ll keep you posted.
Is this closeley related enough to be included in the next sprint also? (Hoping)!
Any updates on this?
I have an object which is non-uniformly scaled at run-time, so I can’t use the scaling at factory creation workaround.
It’s a task that we wish to pick up soon.
It’s been in our sprint a couple of times, but it has been out-prioritized by other more pressing things.
I also look forward to finding a solution to this problem 
A little bit of a status update:
We’re wrapping up the 1.2.150 release now (for Monday). It will contain the synchronous raycast btw!
For 1.2.151 we plan to finish the Facebook upgrade and a bunch of other things (flipbook animation cursors!).
Aaaaaand then for 1.2.152 we plan to do a physics focused sprint where we’ll expose more physics settings and work on things such as physics shape scaling and running the physics simulation in a separate thread.