Re: [PATCH 3/3] gdb/selftest-arch: Make register_test_foreach_arch generate arch tests lazily

[Lancelot SIX <Lancelot.Six@amd.com>]

On 15/04/2022 18:52, Tom Tromey wrote:
> [CAUTION: External Email]
> 
>>>>>> "Lancelot" == Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Lancelot> To fix this, make register_test_foreach_arch register a lazy selftest
> Lancelot> generator.  This way when the test generator is eventually executed, all
> Lancelot> architectures are registered and we do not have a dependency on the
> Lancelot> order the initialize functions are executed in.
> 
> Lancelot> -void
> Lancelot> -register_test_foreach_arch (const std::string &name,
> Lancelot> -                         self_test_foreach_arch_function *function)
> Lancelot> +static std::vector<selftest>
> Lancelot> +foreach_arch_test_generator (const std::string &name,
> Lancelot> +                          self_test_foreach_arch_function *function)
> Lancelot>  {
> 
> Is there some deep reason this has to return a vector?
> It seems like the old code, just calling register_test directly here,
> would also have been fine.
> 
> Tom

Hi,

The reason I went for this is to make the interface clearer about what 
it does.

In the current implementation, I have (in selftest.h):

```
using selftests_generator = std::function<std::vector<selftest> (void)>;
extern void add_lazy_generator (selftests_generator generator);
```

My initial implementation did call `register_test` directly in the 
callback, but then selftest_generator is "just":

```
using selftest_generator = std::function<void (void)>
```

which in my opinion does not communicate much about its intent or use. 
Sure the vector is an overhead, but this machinery is really not on hot 
code path so I did not consider this an issue.

That being said, I am OK with calls `register_test` directly in the 
callback, depending on what the feedbacks are.

Best,
Lancelot.