[Bug 221533] strstr/c23/constness

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=221533

--- Comment #1 from Alejandro Colomar (alx@kernel.org) ---
Hi,

On 2026-05-16T16:47:42+0000, bugzilla-daemon@kernel.org wrote:
> Hi,
>   In C23 strstr can return a const char* if the first parameter is a const
> char*;  this is also true of a bunch of the other routines.
> 
>   I'm not really sure how this should be documented, I guess two entries for
> each function with some note saying it's only C23 and later?

I've been thinking about this.  At the moment, I've documented one
function that has this property: strnul(3) (which is a gnulib extension,
not part of C23).  There, I've used C++-style, documenting it with two
prototypes, as if it were a function overload.  I think we should do the
same for the C23 const-preserving macros.

        SYNOPSIS
             #include <string.h>

             char *strnul(char *s);
             const char *strnul(const char *s);

I'll work on some patches for the C23 APIs.

> See:
>
> https://sourceware.org/cgit/glibc/commit/string/string.h?id=cd748a63ab1a7ae846175c532a3daab341c62690
> 
> for the glibc change that implemented it.
> 
> It can break some existing code, e.g. something like:

Indeed, it broke mutt(1), for example.  That allowed finding a few bugs
in mutt(1), which were modifying read-only strings.  Most of the cases
were false positives, though, which could be fixed easily by adding
const.

> int foo(const char *a)
> {
>    char *bar = strstr(a, "frog");
> }
> 
> gives:
> t.c:9:16: warning: initialization discards ‘const’ qualifier from pointer
> target type [-Wdiscarded-qualifiers]
>     9 |    char *bar = strstr(a, "frog");
>       |                ^~~~~~

Yup.


Have a lovely day!
Alex

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.