Re: [PATCH v6 5/8] GDB/Python: Use None for `var_zuinteger_unlimited' value set to `unlimited'

["Maciej W. Rozycki" <macro@embecosm.com>]

On Mon, 31 Oct 2022, Simon Marchi wrote:

> >> And I do agree that fixing the API once will reduce the long-term costs
> >> for everybody (us and users bumping in the inconsistency and losing
> >> time).  So, I am fine with fixing PARAM_ZUINTEGER_UNLIMITED in this
> >> case.
> > 
> >  Good.  Given that you are in favour to making this change and Andrew was 
> > against, do you think we need another opinion?  Have we reached consensus?
> 
> I didn't know Andrew was against, can you point me to that discussion?

 This is why I went into such a lengthy justification for the change.  I 
didn't want to just sneak the change in given Andrew's previous concern:
<https://sourceware.org/pipermail/gdb-patches/2022-June/190306.html>.

> If the thing is usable as-is, despite being awkward, the alternative is
> to just document as clearly as possible the awkwardness.

 One problem with that is the generalisation I implement in 6/8 (2/4 in 
v7) makes it even more awkward: we do retain PARAM_ZUINTEGER_UNLIMITED in 
the Python interface for compatibility reasons, but it will have to be 
special-cased and extra handling will have to be added (possibly a boolean 
flag driving the handling of the "unlimited" value and translating it to 
-1 regardless of the underlying implementation value) to make it different 
from PARAM_PINTEGER (if we ever expose it to the user) with an "unlimited" 
keyword explicitly added.  Obviously an explicitly added keyword will 
always translate to `None'.

 I didn't want to raise the implementation issue at the beginning as I 
think it's the implementation that should follow the interface and not the 
other way round, but perhaps it's worth noting in this case after all.

  Maciej