Re: [PATCH v2][PR gdb/24052] Implement 'set print zero-values on|off'

[Pedro Alves <palves@redhat.com>]

On 5/31/20 3:58 PM, Hannes Domani via Gdb-patches wrote:
>  Am Sonntag, 31. Mai 2020, 15:39:05 MESZ hat Pedro Alves <palves@redhat.com> Folgendes geschrieben:

>> The thing is that an enum does not measure a quantity or offset.
>> "0" has no usual particular significance compared to
>> other enumerators.  While with pointers and integrals, usually
>> "0" has significance, meaning no quantity, no offset or no object,
>> or in general absence of the property being measured by the variable.
>>
>> For example, here's GDB's auto_boolean:
>>
>> /* * A generic, not quite boolean, enumeration.  This is used for
>>      set/show commands in which the options are on/off/automatic.  */
>> enum auto_boolean
>> {
>>    AUTO_BOOLEAN_TRUE,
>>    AUTO_BOOLEAN_FALSE,
>>    AUTO_BOOLEAN_AUTO
>> };
>>
>> I'd think it confusing that "zero-values off" would hide
>> AUTO_BOOLEAN_TRUE, but not AUTO_BOOLEAN_FALSE.
>>
>> Here:
>>
>> extern enum language_mode
>>    {
>>      language_mode_auto, language_mode_manual
>>    }
>> language_mode;
>>
>> What's the significance of hiding auto but not manual?
>>
>> Here:
>>
>> /* alignment enum */
>> enum ui_align
>>    {
>>      ui_left = -1,
>>      ui_center,
>>      ui_right,
>>      ui_noalign
>>    };
>>
>> Why hide ui_center, instead of the other enumerators?
>>
>> Etc.
> 
> It seems we have very different views about this.
> I don't think it's confusing at all to hide AUTO_BOOLEAN_TRUE/
> language_mode_auto/ui_center in these cases.
> 

OK, if such different views are both reasonable, then this
normally means that the larger set of users will also contain
people with such opposing views, which calls for making it
optional.  Maybe:

 set print zero-values all / non-enums / none

> (For me it's more confusing that AUTO_BOOLEAN_TRUE is first in this enum.)
> 
> If you don't want to hide it, just don't use -zero-values off.
> 

Even the original reporter in the PR suggested only removing
zero enums under an option:

 "Optionally also removing those enums which evaluate to zero would
 save even more unneeded information."

>> (gdb) p g_out
>> $1 = {pad1 = {c = 0 '\000', i = 0}, pad2 = {c = 0 '\000', i = 0}}
>> (gdb) p out
>> $2 = {pad1 = {c = 0 '\000', i = 0}, pad2 = {c = 0 '\000', i = 0}}
>>
>> (gdb) p -zero-values off -- g_out
>> $3 = {}
>> (gdb) p -zero-values off -- out
>> $4 = {pad1 = {}, pad2 = {}}
>>
>> As you see, $3 and $4 gave different outputs, due to the padding.
> 
> I agree that this might be weird, but I kinda see this as a feature.

What's the value of the feature?  I think it's a hard to justify feature,
because garbage in padding happens randomly, and naturally, and doesn't
affect the value at the language level.  I very much question the value
in wanting a different output here, and I hazard a guess that you were
initially surprised with this case too.  IMO it's just a bug not to
consider it.

>> Why print "static_field = 0" when zero-values is off?
> 
> When printing structures, I usually don't care about the static members.
> 
> And with -zero-values off it should just display the parts that have some kind
> of value.
> So now I kinda want to hide all static members when -zero-values off, no
> matter what their real value is.

NAK.  Let's keep options orthogonal.

Thanks,
Pedro Alves