Re: [PATCH 06/13] gdb: check exact results in the canonicalize selftest

[Simon Marchi <simark@simark.ca>]

On 10/21/26 11:21 AM, Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi <simon.marchi@efficios.com> writes:
> 
> Simon> The canonicalize selftest uses should_be_the_same, which canonicalizes
> Simon> two names and checks that the results are equal.  This doesn't check
> Simon> what the canonical form actually is: both names going through the same
> Simon> wrong conversion would still pass.
> 
> I think the idea here is that gdb shouldn't actually care which form is
> the canonical one.  So I'm wondering if there's some reason you think it
> should?

I think it's nice to see how the function actually transforms its
inputs, it helps understand what it does.  And in a sense documents what
the canonical form actually is, which could be useful.

It also makes the test a bit more robust.   If we have:

  should_be_the_same ("something<void ()>", "something<  void()  >");

And I introduce a bug that makes the function "forget" to print "void",
then both sides will canonicalize to:

  something<()>

The test will still pass, when it could have caught the problem.

> Maybe some kind of comment explaining that the rest of gdb ought not
> care would be sufficient.  As in, if some future developer needs to
> change the canonicalizer to return something else for some reason,
> they'd be free to.

I think that the rest of GDB cares to some extent what the canonical
for is.  The functions that work with canonicalized names as inputs need
to know what they could possibly encounter.

With my patch, if a developer needs to change what the canonical form
is, they will need to update the tests accordingly.  But I think that
it's a good thing, because it will make it clear how the behavior
actually changes.

Simon