Re: [myhdl-list] AttributeError in 0.8-dev but not in 0.7

[Christopher F. <chr...@gm...>]

On 1/24/2013 8:08 AM, Christopher Felton wrote:
> On 1/24/2013 7:41 AM, Per Karlsson wrote:
>> The problem is that val is an intbv (that is checked) but next is a bool
>> (and next is never checked). So next has no ._val
>> I don't think the problem can be reproduced using mere assignments. It is,
>> I think, the mixed types in the conditionals that cause the error.
>> I'll make another try at reproducing it 'in vitro' once I get the time.
>> /Per
>>
>> On Thu, Jan 24, 2013 at 1:15 PM, Christopher Felton
>> <chr...@gm...>wrote:
>>
>>> The /self._val._val = next._val/ is correct, it says
>>> assign the /Signal.intbv._val/, the _val of the signal
>>> is an intbv (per the check right before).
>>>
>>
>
> Hmmm, I don't think a conditional will cause the the
> myhdl.simulator to invoke the /update/ functions.
>
> Looking at the code, I could see the attribute hazard in
> the /_update()/ function if:
>
>       _val  : type == intbv
>       _next : type == bool
>
> But for the above to occur the /Signal._SetNextVal/ would
> need to be assigned to /Signal._SetNextBool/.  Now, for that
> to occur the /Signal/ would need to be instantiated with a
> /bool/ ...
>
> Do you see the error now?  You need a complex set of
> events.  You need to create a /Signal/ of type /bool/
> then assign it to and /intbv/ and then assign it to a
> /bool/ again.  If your code inter-mixes /bool/ and
> /intbv[1:]/ a bunch there is this potential to hit this.
>
> code snip:
>
>       x = Signal(bool(0))
>       x.next = intbv(1)[1:]
>       print(type(x._val), type(x._next))
>       x._update()
>       print(type(x._val), type(x._next))
>       x.next = False
>       print(type(x._val), type(x._next))
>       x._update()
>
>
> The above will cause the error you are seeing.
>
> Now the question is, does the design intend for /bool/
> and /intbv[1:]/ interchanged.  If it does then this needs
> to be fixed, if it doesn't then we need to add an assert
> that prevents the first /intbv/ assignment to the bool.
>
> Regards,
> Chris Felton
>
>

I did not test the above with 0.7 as you indicated, you only
observed the issue with 0.8-dev.  I can diff the version and
see what might have changed (at some point).

Regards,
Chris