WER dumps generate corrupted stacks

[qmuntal]

The Core OS team are trying to use WER dumps generated by setting GOTRACEBACK=wer. They can successfully generate dumps for crashing programs, but with what seems to be corrupted stack traces.

WER support for Go was upstreamed by us in Go 1.21 (see golang/go#57441). One of the limitations at that time is the lack of test coverage related to examining the dump content using a debugger, as this is out of the Go standard library test suite. This means that something might have been broken in the meantime. We tried to partially mitigate this lack of upstream testing by testing it ourselves in https://github.com/microsoft/go-pdb/blob/fa4e71cc7269ca83f021f07948e49b1f7af7eae1/gopdb/gopdb_windows_test.go#L209, but the test suite is not extensive.

Back to the issue, it can be reproduced by a simple hello world crashing app:

package main

func main() {
    panic("oh no!")
}

And then using Delve or WinDbg to examine the dump.

Reported by @kevpar and @mtbar131

[qmuntal]

While investigating this issue I found that the exception address from the exception record was not correct in case of a panic. I've submitted an upstream CL to fix this: https://go-review.googlesource.com/c/go/+/619296, and it contains tests!

Note that it doesn't fix the corrupted stack trace, but at least WinDbg is able to show the faulting source code: