Re: [PATCH v6 06/18] libxl: write qemu arguments into separate xenstore keys

[Ian Jackson <ian.jackson@xxxxxxxxxx>]

Jason Andryuk writes ("Re: [PATCH v6 06/18] libxl: write qemu arguments into 
separate xenstore keys"):
> On Mon, May 18, 2020 at 10:20 AM Ian Jackson <ian.jackson@xxxxxxxxxx> wrote:
> > I think this should be "goto out".  That conforms to standard libxl
> > error handling discipline and avoids future leak bugs etc.
> > libxl__xs_transaction_abort is a no-op with t==NULL.
> >
> > Also, it is not clear to me why you chose to put this outside the
> > transaction loop.  Can you either put it inside the transaction loop,
> > or produce an explanation as to why this approach is race-free...
> 
> I just matched the old code's transaction only around the write.  "vm"
> shouldn't change during runtime, but I can make the changes as you
> suggest.

Ah I see.  I hadn't spotted this duplication.

As there is only one caller of libxl__write_stub_dmargs the messing
about with %d/vm and the transaction and so on could be factored out
and only the actual arg processing made conditional.

I would prefer that, to be honest.  But I don't want to derail this
series at this point by asking you to take on refactorings that I
ought to have asked for sooner.

So I'll take it if you make the new code the way I like it, as I
suggest above.  Maybe it will be refactored later (perhaps even by
me...)

Regards,
Ian.