The linux kernel doesn't guarantee that the data written to a pty will
instantly be available for reading on the other side. So what sometimes
happens here is that the inferior dumps all register values and exits,
lldb registers the exit, goes to read the inferior stdio, but the data
is not there.
Depending on how the machine is configured, the frequency of this
occurring can range from "almost never" to "pretty much always". This
patch side-steps the issue by redirecting the inferior output to a file.
Files have better synchronization properties (plenty of applications
depend on those), and this fixes the flakyness of these tests on
affected machines.
Any reason not to use %T directly in paths rather than changing working directory? I suppose either way is fine but I have a bad feeling that this could be confusing in the long run.