Re: [PATCH] Fix -Wshadow=local warnings in elfos.h

[Joseph Myers <joseph at codesourcery dot com>]

On Sat, 5 Oct 2019, Bernd Edlinger wrote:

> > Like
> > 
> > #define DEFAULT_SOME_MACRO(PARMS) { lots of code }
> > 
> > becomes
> > 
> > #define DEFAULT_SOME_MACRO(PARMS) default_some_macro(PARMS)
> > 
> > static inline int
> > default_some_macro (int parm, long another)
> > {
> >   lots of code;
> > }
> > 
> > The point is that all this is completely *local* changes.
> > 
> 
> Hmm, I tried this but it does not work:

The answer to that is an out-of-line function.

See the default_register_move_cost function in targhooks.c, the 
register_move_cost hook and the REGISTER_MOVE_COST target macro, for 
example.  There is a strategy for incremental conversion of a single 
target macro to a hook that goes as follows:

* Define a corresponding target hook.  Make the default definition in 
targhooks.c be one that tests whether the target macro is defined and 
either uses that macro or the existing default as appropriate.  Make 
tm.texi.in document the hook alongside the macro, saying that the macro is 
obsolete and new targets should use the hook.

* Convert all *uses* of the target macro to use the hook instead.  There 
is no need to change any of the definitions in the various targets with 
non-default definitions.

* Remove the default definition of the target macro, as no longer needed 
now targhooks.c tests whether the target macro is defined or not.

At that point, you have a partial conversion of that target macro to a 
hook (and consequent gains regarding avoiding shadowing) without having 
touched any of the target-specific code.  Any target can safely convert 
its own definition of the macro to a definition of the hook, independently 
of all other targets and all other hooks.  Once all targets have changed, 
the macro can be poisoned and the documentation of the macro and support 
in targhooks.c for using the macro removed, in the usual fashion for 
removing a target macro - but that might happen a long time afterwards.  
It's more work than a pure local conversion to an inline function, or than 
just renaming variables, but significantly less than a full conversion of 
the macro to a hook (and in particular should not need testing on more 
than one target).

-- 
Joseph S. Myers
joseph@codesourcery.com