Re: [PATCH v2] Update the 'g' packet documentation

[Pedro Alves <pedro@palves.net>]

On 2023-01-30 9:15 p.m., Tom Tromey wrote:
>>>>>> "Pedro" == Pedro Alves <pedro@palves.net> writes:
> 
> Pedro> Here, the new text still uses "collected", but lost the reference to trace frames.
> Pedro> It seems to me that that will result in people not knowing what "collected"
> Pedro> means in this context.
> 
>>> Yeah, I wanted to get rid of the trace frame note, because it's
>>> confusing -- 'x' can be sent any time, not just a trace frame.
> 
> Pedro> Yeah, but then people won't know what "collected" here means.  Also, in the normal
> Pedro> case you shouldn't really end up with unavailable registers -- if some register
> Pedro> really doesn't exist, then the target description should ideally not describe it.
> 
> While I agree with that, it's also the case that tdep code can reject
> such a description, and we ran into a case like this in the wild --
> where a gdbserver does not report a register, causing i386-tdep to
> reject the description.  However, at one point AdaCore had a port for
> this target, and that port reported the register but sent back 'x' --
> which worked.

Curious.  It sounds like we made some register in a tdesc feature mandatory,
even after the tdesc feature existed in previous releases?  That shouldn't have
happened.

> 
> Also, I wanted to say, I think I misread your first reply.  If I'd read
> it correctly I would have rewritten the text.  I'm sorry about that.

No worries at all.

> Pedro> How about this version of the patch?  It combines your original patch with
> Pedro> my suggestions, and also extends it further to describe the normal live target
> Pedro> scenario and include a remark about just not including the register in the tdesc.
> Pedro> (Note: I did not add an xref for the target description section at the end because
> Pedro> there's already one in the preceding paragraph.)
> 
> This sounds great.  Thank you.

Great, I'm going to merge it.

Thanks,
Pedro Alves