Re: [PATCH] Teach "info breakpoints" to show the address of a -location watchpoint

[Pedro Alves <palves at redhat dot com>]

On 03/30/2016 01:48 AM, Patrick Palka wrote:
> On Tue, Mar 29, 2016 at 3:40 PM, Pedro Alves <palves@redhat.com> wrote:
>> On 03/26/2016 05:34 PM, Patrick Palka wrote:
>>> Currently the "info breakpoints" command leaves empty the Address column
>>> corresponding to a -location watchpoint, even though there is an address
>>> internally tied to this watchpoint.  Instead of printing nothing, this
>>> patch makes the type and the computed address of the -location
>>> watchpoint get printed.  Conviently the exp_string_reparse already has a
>>> pretty-printed string that contains this information.
>>>
>>> The new output of "info breakpoints" looks something like:
>>>
>>> Num     Type           Disp Enb Address            What
>>> 2       hw watchpoint  keep y   (tree_code *) 0x00007ffff7ff4990 -location decl.base.code
>>>
>>
>> Hmm, isn't this going to look more awkward the longer the type name is?
>> Do we include C++ namespaces in there as well?  Usually, the column
>> that has variable width is the last one.
>>
>> If you're looking for the width of the watched range, a shorter alternative
>> would be to not print the type, but instead use ADDR:LEN notation, like:
>>
>> Num     Type           Disp Enb Address            What
>> 2       hw watchpoint  keep y   0x00007ffff7ff4990:4 -location decl.base.code
>>
>> WDYT?
> 
> If only the last column should have a variable length, then what about
> appending the type and address to the last column, so that the output
> looks like:
> 
> Num     Type           Disp Enb Address            What
> 2       hw watchpoint  keep y                      -location
> decl.base.code at (tree_code *) 0x00007ffff7ff4990
> 
> or what about recursing through component accesses and printing the
> location of the base object, like so
> 
> Num     Type           Disp Enb Address            What
> 2       hw watchpoint  keep y                      -location
> decl.base.code at &((some_struct *) 0x00007ffff7ff4900).base.code
> 
> The latter output would give us the most information, including
> bitfield information (if e.g. the "code" field is a bitfield).

In the latter case, don't you want to see the final resolved address,
and the type of "code"?  Like, assuming type int, and offset from
base by 0x10 bytes:

Num Type           Disp  Enb  Address     What
2   hw watchpoint  keep  y                -location decl.base.code at (int *) 0x00007ffff7ff4910

Or, going the direction of showing the whole path, we could have everything in, by 
showing an (int *) cast before the "&", and using the "Address" column for the final
resolved address, like:

Num     Type           Disp Enb Address              What
2       hw watchpoint  keep y   0x00007ffff7ff4910  -location decl.base.code at (int *) &((some_struct *) 0x00007ffff7ff4900).base.code

And using the ADDR:LEN notation idea:

Num     Type           Disp Enb Address              What
2       hw watchpoint  keep y   0x00007ffff7ff4910:4 -location decl.base.code at (int *) &((some_struct *) 0x00007ffff7ff4900).base.code


Hard to judge what's best without trying it out for a while.

Thanks,
Pedro Alves