From: Boris Gjenero <boris.gjenero_at_gmail.com>
Date: Wed, 21 Dec 2011 13:53:06 -0500

With -Wmissing-declarations, gcc can warn about non-static global
functions and data without a previous declaration. What do others think
about the idea of enabling the warning for the core, after some cleanup?
I examined warnings about stuff in the core, and this is what I found:

* Missing static. Sometimes a function is only static in some
configurations.

* Missing includes. In these cases, nothing ensures that the function
matches the prototype. Also, attributes (such as INIT_ATTR) which may be
found in the prototype don't apply to the function.

* A function that's not used in the current configuration. The correct
header file is included, but preprocessor conditionals exclude the
prototype. Sometimes, it's a function which does nothing and exists for
a feature that isn't implemented for that target. I'm not sure if those
need conditionals or deletion.

* Functions used in other files, where the prototypes are only in C
files using the functions. I think this is undesirable. It defeats the
protection offered by prototypes.

* Totally unused code

* Interrupt handlers and other code only referenced from assembler,
where prototypes are not helpful. This would require a prototype just to
avoid the warning. There's a lot of this in Rockbox, and it's the main
disadvantage of -Wmissing-declarations.

I did not examine code outside of core because there are far more
warnings there. In some cases it's code from other projects, where
changes could make backporting of fixes harder. For these reasons, I'm
only suggesting -Wmissing-declarations for the core.

Regards,

Boris
Received on 2011-12-21