Tagging subscribers to this area: @dotnet/area-system-text-json, @gregsdennis
See info in area-owners.md if you want to be subscribed.

Where was the stack overflow occurring?

Where was the stack overflow occurring?

The DeepEquals method itself.

Where was the stack overflow occurring?

The DeepEquals method itself.

I'd asked about stack overflows here:
#104579 (comment)
and the response was that wasn't a concern because it would fail before that in the JsonDocument parser. But that's not the case?

I'd asked about stack overflows here: #104579 (comment) and the response was that wasn't a concern because it would fail before that in the JsonDocument parser. But that's not the case?

That's still the case, however if you configure the max permitted depth to be something ridiculous (like this test is doing) you're opening yourself up to instances that can stack overflow. The reason why I hadn't caught this SO at the time was that it isn't actually occurring if you run dotnet test, it only impacts xunit.console hosts.

I'd asked about stack overflows here: #104579 (comment) and the response was that wasn't a concern because it would fail before that in the JsonDocument parser. But that's not the case?

That's still the case, however if you configure the max permitted depth to be something ridiculous (like this test is doing) you're opening yourself up to instances that can stack overflow. The reason why I hadn't caught this SO at the time was that it isn't actually occurring if you run dotnet test, it only impacts xunit.console hosts.

That max depth is being passed to the parser. Is it not using recursion? Or are its stack frames smaller and so there's a range where it wouldn't fail but equals would?

The JsonDocument parser isn't recursive as such, so it can theoretically support arbitrary depths.

Should I not be concerned then that you could be handed an arbitrary JsonElement, call DeepEquals, and crash due to stack overflow?

The algorithm could be made iterative at the expense of allocating a stack, but it doesn't seem particularly worthwhile to me. JsonSerializer is known to stack overflow if the MaxDepth is set to something large and cycles are encountered.

/ba-g unrelated test failure, merging to fix crashing test.