Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20150417193320.GA11349@wilbur.25thandClement.com>
Date: Fri, 17 Apr 2015 12:33:20 -0700
From: William Ahern <william@...handClement.com>
To: musl@...ts.openwall.com
Subject: Re: Explicit casts in ctype.h suppress compiler warnings

On Fri, Apr 17, 2015 at 12:52:38PM -0400, Rich Felker wrote:
> On Fri, Apr 17, 2015 at 06:49:54PM +0200, Jens Gustedt wrote:
<snip>
> > I generally think that casts are a bad idea, anyhow, and should only
> > be used where it must be done, that is basically for pointer to
> > integer conversion (and back). Code like this
> > 
> > #define isdigit(a) (((unsigned)(a)-'0') < 10)
> > 
> > can easily be replaced by
> > 
> > #define isdigit(a) (((unsigned const){a}-'0') < 10)
> > 
> > to change the explicit conversion to an implicit one in the
> > initializer of the compound literal. Then, any compiler would have to
> > diagnose if "a" would be a pointer.
> 
> In another place (math.h) I removed this type of compound literal
> usage because it was incompatible with C++, but the macros are
> suppressed in C++ anyway. Still they might break -pedantic with
> -std=c89. I do like this approach best in principle if it works
> though, because the rules for when an error occurs are basically the
> same as the rules for a real function.

I think C++11 brace initialization

	int { 42 }

works nearly identically to C99 compound literal definition

	(int){ 42 }

The big difference is that the lifetime of C++ temporaries have expression
scope, whereas compound literals have block scope. But that's irrelevant
where the values will be copied and no pointer is derived.*

A simple macro could be used to select the syntax. However, I don't think
something like `unsigned char { 42 }' will work. The type name needs to be a
single identifier, so a typedef would be needed: u_char { 42 }.

* Unfortunately, by default g++ accepts C99-style compound literals, but for
whatever reason gives them expression-scoped lifetimes as-if they were C++
temporaries. This gave me and some other people grief when when using
pointers to compound literals, either explicitly or implicitly through
array-to-pointer decay). See Bug #53220. Everything seemed to work when
compiled as C++, unless you were fortunate enough to get a crash near the
offending code. More recent versions of g++ now at least warn when a pointer
is derived from a compound literal.

I think clang++ has the same behavior.

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.