[m-rev.] for post-commit review: infrastructure for testing new filename translation methods

[Julien Fischer]

On Sun, 4 Jun 2023, Zoltan Somogyi wrote:

> Add infrastructure for testing filename generation.
> 
> compiler/file_names.m:
>     Add a mechanism for comparing the outputs of two different implementations
>     of the conversion from a module_name/extension pair to a file_name,
>     for all the extensions we use in such translations.
>
>     The intention is that this be used to compare a proposed replacement
>     of module_name_to_file_name_ext with module_name_to_file_name_ext itself.
>     For now, the comparison is a dummy, comparing module_name_to_file_name_ext
>     with itself.
>
>     Delete a reference to IL code.
> 
> compiler/mercury_compile_main.m:
>     Invoke the new code in file_names.m if an experiment option is set.
>
>     Add an XXX.
> 
> diff --git a/compiler/file_names.m b/compiler/file_names.m
> index 316d8f8fc..96b9e6d5e 100644
> --- a/compiler/file_names.m
> +++ b/compiler/file_names.m
...

> +:- func string_extensions = list(string).
> +
> +% We don't test the .m extension, since that is handled separately
> +% by module_name_to_source_file_name.
> +string_extensions = 
> +    ["",
> +    ".$O",

$O will expand to .o or .obj depending on the underlying object file extension
used by the platform ...

> +    ".mode_constraints",
> +    ".module_dep",
> +    ".o",

... so why is .o listed separately here?  (Or why is .obj not also listed?)

> +    ".opt",
> +    ".optdate",
> +    ".opts",
> +    ".order",
> +    ".order-trans-opt",
> +    ".pic_o",
> +    ".prof",
> +    ".realclean",
> +    ".request",
> +    ".so",

".dylib" (the shared libray extension on macOS) is missing.

Julien.