Re: Race with atexit functions in system emulation

[Pavel Dovgalyuk]

On 02.07.2020 10:49, Alex Bennée wrote:
> Pavel Dovgalyuk <pavel.dovgaluk@gmail.com> writes:
>
> > Is it true, that semihosting can be used to access (read and write) host
> > files from the guest?
>
> They can - but in these test cases we are only using semihosting for
> console output and signalling an exit code at the end of the test. I
> don't think that gets in the way of record/replay (aside from the exit
> race described).

Ok.

> > In such a case it can't be used with RR for the following reasons:
> > 1. We don't preserve modified files, therefore the execution result may
> > change in the future runs.
> > 2. Even in the case, when all the files are read only, semihosting FDs
> > can't be saved, therefore it may not be used with reverse debugging.
>
> This raises a wider question of what is the best way to indicate support
> (or lack of support) for a particular device to a user? Do we need a
> "replay aware" list or annotation?

There is a replay blocker, which is set at the command-line parsing phase.
E.g., user gets an error when tries to enable -smp >1.


Pavel Dovgalyuk