Re: [PATCH 5/6] gdbsupport/selftest: print operand values when a SELF_CHECK fails

[Simon Marchi <simark@simark.ca>]

On 10/27/26 10:13 AM, Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi <simon.marchi@efficios.com> writes:
> 
> Simon> Some previously used constructs are now rejected at compile time, and
> Simon> require a few changes in the tests:
> 
> Simon>  - A top-level && or ||.
> 
> I think this approach is very clever but I'm not really a fan of the result.
> It seems completely reasonable to me to want to write || or && in a test.

If you really need it (you only really "need" it for ||), you can always
add extra parentheses.

> Simon>  - A lone non-bool value, like SELF_CHECK (ptr).
> 
> This isn't really that weird either.

We could make it work, I just chose not to because of our coding style.

> Simon> Operands are printed using a value_to_string function that is a long
> Simon> series of "if constexpr" to print various types of operands.  It ends
> Simon> with an always-false static_assert, such that trying to print an
> Simon> unsupported type will cause a build failure, and will require telling
> Simon> the function how to print said type.
> 
> Why this way and not overloads?

I guess you mean templates with SFINAE to select the right printer?  If
so: because it's easier to understand than separate templated functions,
especially the order in which rules are applied.  With templates, you
have to know the template deduction rules, which isn't obvious.  With a
single function, the decision tree is right there, easy to follow.

> It seems to me that the drawback of this approach is that now the test
> framework has to have visibility on every possible thing that could be
> printed.

We can abstract it somewhat by having the test framework just call
`val.to_string()` or `to_string(val)`, and then the specific `to_string`
implementations are not known by the test framework.

Simon