Re: mdocml: To remove the const qualifier from a pointer to an object

[Joerg Sonnenberger <joerg@britannica.bec.de>]

On Sat, Jul 16, 2016 at 12:08:46AM +0200, Ingo Schwarze wrote:
> > The macro exists
> 
> The macro does not exist.  Neither the C standard nor POSIX specify
> it or even indicate that something like that might be needed or
> even useful.  It was a purely local quirk in one single place of
> the mandoc codebase.  Besides, i checked with Philip Guenther, who
> is both an expert in C and in POSIX, that no contortions are needed
> here.

This doesn't make any sense at all given that you just removed the
macro.

> > for two reasons:
> > (1) It makes auditing easier by giving something simple to look for.
> 
> It makes auditing harder due to obfuscation.  What a plain cast
> does is obvious on first sight.  A macro needs to be looked up when
> auditing the code to understand what it does, and due to the
> unnecessary quadruple cast const char **" -> const void * ->
> intptr_t -> void * -> char * const * instead of one single cast
> it makes auditing even harder.

This is plainly wrong. A type cast can have a hundred different
meanings. It is far from obvious that it is only meant to remove a type
qualifier. If you can't derive the intention from a macro called
UNCONST...

> I see no point in specifically auditing for casts removing const.

Given that const memory can actually be read-only, removing the logical
property is a clear sign of either a misdesigned interface, a workaround
for a language defect or a plain bug to be waiting. It is no less a
warning sign than use of strcpy..

> > (2) The macro actually stopped -Wcast warnings.
> 
> You mean, -Wcast-qual?  That's not enabled by default for a reason:
> It's not useful in production builds.  (It may occasionally be
> helpful when doing a one-time audit of a dubious code base to
> identify candidates for erroneous casts, though.)

It is used in production builds by many systems, including FreeBSD and
NetBSD, for a lot of software. Just because you don't do that in OpenBSD
doesn't mean it doesn't create problems for others.

> > The replacement does not.
> 
> Just don't enable -Wcast-qual for compiling mandoc.  I consider
> that option harmful because it drives developers toward obfuscation.

I beg to differ and a lot of people agree with me.

Joerg
--
 To unsubscribe send an email to tech+unsubscribe@mdocml.bsd.lv