Re: Preview: portable dumper

[Daniel Colascione]

On 02/16/2018 01:09 PM, Paul Eggert wrote:
> On 02/16/2018 12:43 PM, Daniel Colascione wrote:
> > IMHO, we should enable -Wconversion more broadly.
>
> My experience is just the opposite: i.e., that -Wconversion causes more 
> trouble than it cures. The pdumper change to INTEGER_TO_CONS is an 
> example of trouble. The only reason for that change is to work around a 
> compiler bug in GCC that is caused by -Wconversion, a bug that leads to 
> a false alarm. I suggest at least the attached patch, which limits the 
> damage to pdumper.c instead of letting it spread to other Emacs modules. 
> But better yet, I suggest dropping the idea of using -Wconversion even 
> on pdumper.c, as it's counterproductive and in high-quality code almost 
> inevitably leads to further obfuscation like ALLOW_IMPLICIT_CONVERSION 
> and DISALLOW_IMPLICIT_CONVERSION.

-Wconversion helped me find real bugs and otherwise mysterious bugs in 
pdumper. It also, apparently, found a longstanding subtle bug in a 
completely different part of Emacs. Outside of certain specialized 
casting facilities, -Wconversion ought to warn only on things that are 
actually problems. Have a counter-example?

The ALLOW_IMPLICIT_CONVERSION and DISALLOW_IMPLICIT_CONVERSION macros 
are there in pdumper to account for the rest of Emacs not being 
conversion-clean; if everything worked with -Wconversion, we wouldn't 
need the bracketing.

> PS. Double-parens like that shouldn't be needed in macro bodies, as each 
> macro must parenthesize its argument properly anyway.

Thanks.