Re: [PATCH v2 1/5] GDB: Ignore `max-value-size' setting with value history accesses

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

On Fri, 13 Jan 2023, Andrew Burgess wrote:

> > @@ -1752,12 +1763,14 @@ value_release_to_mark (const struct valu
> >    return result;
> >  }
> >  
> > -/* Return a copy of the value ARG.
> > -   It contains the same contents, for same memory address,
> > -   but it's a different block of storage.  */
> > +/* Return a copy of the value ARG.  It contains the same contents,
> > +   for the same memory address, but it's a different block of storage.
> > +   If CHECK_SIZE is true, then throw an exception whenever the size
> > +   of memory allocated for the contents of the value would exceed
> > +   max-value-size.  */
> >  
> > -struct value *
> > -value_copy (const value *arg)
> > +static struct value *
> > +value_copy (const value *arg, bool check_size)
> >  {
> >    struct type *encl_type = value_enclosing_type (arg);
> >    struct value *val;
> > @@ -1765,7 +1778,7 @@ value_copy (const value *arg)
> >    if (value_lazy (arg))
> >      val = allocate_value_lazy (encl_type);
> >    else
> > -    val = allocate_value (encl_type);
> > +    val = allocate_value (encl_type, check_size);
> 
> I wonder, maybe value_copy should never check the max-value-size.  As
> you point out above, the max-value-size was introduced to catch cases
> where attempting to read a value from the inferior would cause GDB to
> try an allocate a stupid amount of memory.
> 
> We don't currently have any mechanism in GDB to try an cap the
> cumulative memory usage across all values, which suggests that
> currently, once a value has been read into GDB we assume we can safely
> make as many copies as we want.
> 
> And so, is there any reason why value_copy shouldn't always disable the
> size check?

 I had to investigate that and found out that in addition to value history 
accesses `value_copy' is only used for internal variables (which is fine; 
I guess that's one reason for the minimum of 16 set for `max-value-size'), 
and then an obscure piece of code called `make_cv_value'.

 Said function is only referred to from Python and Scheme support code, 
and its interaction with `max-value-size' is not covered by the testsuite.  
I have added a suitable test case then and that has verified it is indeed 
fine for `value_copy' to always disable the size check.

 Thank you for the suggestion then.  Since I missed a couple of copyright 
year updates with the original submission and also Tom's recent conversion 
requires the use of `allow_ada_tests' rather than `skip_ada_tests' in Ada 
test cases, I'll respin the patch series with all the necessary updates.

  Maciej