On Tue, Nov 8, 2016 at 2:01 PM, Ulrich Weigand <uweigand@de.ibm.com> wrote:
> Yao Qi wrote:
>
>> If we go to the direction you suggested, val_print routines shouldn't
>> assume that VAL contents are already fetched.  Before we do printing,
>> optimized-out checking, or available checking, we should make sure VAL
>> is not lazy.  Nowadays, we can't fetch the value partially, because
>> 'lazy' is a bool attribute, we either fetched all contents or haven't
>> fetch all contents.
>
> Right.  But if we ever went to actually create sub-value objects, then
> we might be able to take advantage of only fetching the contents for
> those sub-objects are actually printed.  In any case, I still think
> this new patch is preferable over calling value_fetch_lazy all the
> time ...
>
>> How is the patch below?  I don't write the ChangeLog entry yet.
>> Regression tested on x86_64-linux.
>
> Looks good to me.

Thanks, I'll push it in.

>
>> --- a/gdb/extension.c
>> +++ b/gdb/extension.c
>> @@ -478,9 +478,9 @@ free_ext_lang_type_printers (struct ext_lang_type_print=
>> ers *printers)
>>    xfree (printers);
>>  }
>>  =0C
>> -/* Try to pretty-print a value of type TYPE located at VALADDR
>> -   + EMBEDDED_OFFSET, which came from the inferior at address ADDRESS
>> -   + EMBEDDED_OFFSET, onto stdio stream STREAM according to OPTIONS.
>> +/* Try to pretty-print a value of type TYPE located at VAL's contents
>> +   buffer + EMBEDDED_OFFSET, which came from the inferior at address
>> +   ADDRESS + EMBEDDED_OFFSET, onto stdio stream STREAM according to OPTION=
>> S.
>>     VAL is the whole object that came from ADDRESS.  VALADDR must point to
>>     the head of VAL's contents buffer.
>>     Returns non-zero if the value was successfully pretty-printed.
>
> This changes the comment but not the actual implementation, this seems
> to be an oversight?
>

It is a mistake in patch splitting.  It is removed from this patch.  Patch
attached is pushed in, with ChangeLog entry.

-- 
Yao (齐尧)