Code style: decide struct initialization style

[iskra-anl]

In GitLab by @perarnau on Jan 29, 2020, 16:12

The library code is full of struct initialization statements, most notably for operators. I've noticed that there are multiple code styles used in the code, with some slight side-effects:

struct aml_struct_ops aml_specific_struct_ops = {
     .fieldname = aml_specific_struct_fieldname,
};

and

struct aml_struct_ops aml_specific_struct_ops = {
     aml_specific_struct_fieldname,
};

I would like to point out that these two styles are not equivalent: the first one allows for partial initialization of the structure, the rest of the fields having, I think, NULL as a default value, while the second one will not compile if any field is missing.

My guess is that each style might make sense for some cases, but we should make sure that the choice was made on purpose.

[iskra-anl]

In GitLab by @perarnau on Jan 29, 2020, 16:12

@Kerilk and @NicolasDenoyelle , fyi.

[iskra-anl]

In GitLab by @NicolasDenoyelle on Jan 30, 2020, 07:43

The more explicit, the better imo.
I would chose first style by default and use the second when it is necessary.

[iskra-anl]

In GitLab by @perarnau on Jan 30, 2020, 09:12

when necessary is the big question: the first style will not fail at compile time if we forget to initialize a field in the structure. Ideally the unit tests would be enough to catch something like that, but I don't know if that's enough.

[iskra-anl]

In GitLab by @perarnau on Apr 6, 2020, 15:15

FYI, clang-format will format both of those differently, in particular it will respect the one line per function style only if the fields are named...

Related to !123

[iskra-anl]

In GitLab by @perarnau on Apr 6, 2020, 15:22

And apparently, the first one is called designated initializer and is a C99 feature, while the other is older.

I think I can be okay with the first style if we make sure to test properly all the methods that are mandatory.

[iskra-anl]

In GitLab by @Kerilk on Apr 6, 2020, 16:05

Especially designated initializer is not completely c++ compliant. This might not be important for source code, but for headers this could become critical.

[iskra-anl]

In GitLab by @perarnau on Apr 6, 2020, 17:33

We only use that on the .c side for static initialization of function pointer structs, so I think we're safe.