Failures on [dds/dap] Handle Sentinels in invoke toString()/getObject and return error to clients about sdk

[derekxu16]

@DanTup and @bkonyi can you please take a look?

[bkonyi]

I'll see if I can reproduce.

[bkonyi]

I can reproduce and the assert is failing with offset == 0. I'm not sure the assert we're hitting is quite right. We check to see if offset < 0 in GetObject and then immediately assert that offset > 0.

[DanTup]

@bkonyi is this something wrong in the VM, or in DAP? The noted change did add a new test that will trigger a bunch of getObject() calls for items in a large list (and will deliberately exhaust the ring buffer to test handling of expired references).

Edit: I re-read it, and guess this is a VM issue? Even if the request to the VM service is invalid, I think it shouldn't fail an assert like this?

[bkonyi]

Yeah, this is a VM issue. We just had a bad assertion in the C++ code that fails if you try and provide an offset or count of 0 to the getObject call.

[DanTup]

We just had a bad assertion in the C++ code that fails if you try and provide an offset or count of 0 to the getObject call.

I presume this didn't show up for me because I was using a bleeding-edge SDK build (which I think doesn't include asserts)? But why didn't it show up on the trybots?

(I'm also surprised we've never sent an offset of 0 in any of the other tests! Was there some other condition, or should offset=0 have failed consistently? 🤔)

[bkonyi]

We just had a bad assertion in the C++ code that fails if you try and provide an offset or count of 0 to the getObject call.

I presume this didn't show up for me because I was using a bleeding-edge SDK build (which I think doesn't include asserts)? But why didn't it show up on the trybots?

Correct, this assertion would only be hit in VMs built in DEBUG mode. It looks like there's a slight hole in the CQ configuration selection. I think we only run analysis and the pkg bots for pkg/dds CQ runs, which don't include any debug configurations.

(I'm also surprised we've never sent an offset of 0 in any of the other tests! Was there some other condition, or should offset=0 have failed consistently? 🤔)

Apparently not! It should have failed consistently in DEBUG builds.