Re: [RFC PATCH] Read pseudo registers from frame instead of regcache

[Simon Marchi <simon dot marchi at polymtl dot ca>]

On 2018-05-26 23:42, Tom Tromey wrote:
> > > > > > "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:
>
> Simon> The problem: Reading pseudo registers from an upper stack frame 
> does not
> Simon> work.  The raw registers needed to compose the pseudo registers 
> are
> Simon> always read from the current thread's regcache, which is 
> effectively
> Simon> frame #0's registers.
> [...]
> Simon> The gdbarch hooks
>
> Simon>   - pseudo_register_read
> Simon>   - pseudo_register_read_value
> Simon>   - pseudo_register_write
>
> Simon> receive a regcache as a parameter, as a way to get the values of 
> the raw
> Simon> registers required to compute the pseudo register value.  
> However, the
> Simon> regcache object always contains the current register values, not 
> the
> Simon> values unwound to the frame we're interested in.
>
> I don't know this area very well, so apologies if this is really off
> target; but I am wondering why not just make a regcache for the desired
> frame and pass that to the gdbarch hooks?

That sounds like a good idea, at least to avoid changing the gdbarch 
interface.  The current regcache always fetches raw registers using 
target_fetch_registers (which gets the current values of the thread's 
registers), but we could have another kind of regcache that instead gets 
raw register values by unwinding from the previous frame.

> Another thing I am curious about is how far back this bug goes.  Like,
> if it were introduced in one of the last few releases, then it might be
> worth bisecting to find out what happened.

The oldest version of GDB with which I could print $ebx is 7.2 (did 
pseudo registers not exist before?), and it exhibits the same problem.  
So I would think that the problem exists since "forever".

Simon