[Lldb-commits] [lldb] [lldb] Expose structured command diagnostics via the SBAPI. (PR #112109)

[Jordan Rupprecht via lldb-commits]

rupprecht wrote:

The `command-expr-diagnostics.test` test added here (migrated from API->shell, really) is flaky for me. It expects this:

```
(lldb) p a+b
         ˄ ˄
         │ ╰─ error: use of undeclared identifier 'b'
         ╰─ error: use of undeclared identifier 'a'
```

But half the time I see this:

```
(lldb) p a+b
           ˄
           │error: use of undeclared identifier 'a'
           ╰─ error: use of undeclared identifier 'b'
```

I haven't dug into why this is the case. One strange thing is that the brokenness is consistent within an LLDB invocation; i.e. running `lldb -o "p a+b"` has a 50% success rate, but running `p a+b` 10 times within a single LLDB session will either work 10 times or fail 10 times. The non-determinism must be something tied to global state/cached/etc.

Has anyone else noticed this? Any ideas what might be going on?

https://github.com/llvm/llvm-project/pull/112109