Re: RFC: Runtime checking of port handling

[Justus Winter]

Hi,

Quoting Samuel Thibault (2016-06-04 17:32:06)
> Justus Winter, on Sat 04 Jun 2016 15:45:15 +0200, wrote:
> > tl;dr: Compiler-assisted runtime checking of port handling in
> > variables with automatic storage duration.  Do we want to go there?
> 
> That's interesting, but the proposed way would need annotating the whole
> source code, I'm afraid we'll make mistakes there.

Are you afraid that it is too much work, or that the process is
error-prone?

On the former point, this method can be applied on a per-function, and
even per-variable basis, so it can be introduced incrementally.  On
the latter, it makes all port handling errors fatal, while providing
us with meaningful error messages and stack traces.

> Passing a port to a function that will deallocate it will probably
> need another macro BTW.

I have provided Mach_port_move for that.

> Ideally we'd have a static analysis tool which knows about the mig rules
> for port references. Perhaps LeakSanitizer could be taught that? (we
> probably want to port it to Hurd, anyway, and it seems very portable)

Yes, static analysis tools are nicer (but come with their own set of
limitations).  Note that the *Sanitizers are also runtime tools.  In
any case, I doubt that we can create a tool that works entirely
without modifying / annotating existing code.

Justus