Re: [9front] strndup should use strnlen and memcpy

["Devon H. O'Dell" <devon.odell@gmail.com>]

Hi Sigrid,

The C2x working draft[1] incorporates strdup and strndup. Of strndup,
it says, "[t]he strndup function creates a string initialized with no
more than size initial characters of the *array pointed to by s*..."
(emphasis mine). Strings in the standard library are always called
strings. While any object occupying fewer than `size+1` bytes must be
a string, this is a side-effect of how the API is specified to behave.

POSIX is ambiguous, but implies that it expects a string either way:
"[t]he strndup() function shall be equivalent to the strdup()
function...", without subsequently placing exceptions on the input
constraints. Many (most?) implementations don't bother to check (or
explicitly allow non-strings), and WG14 likely recognized this when
standardizing the functionality in libc.

Anyway, if taking the "find length before allocation" approach to
implementation, the WG14 wording requires use of strnlen semantics.
One can implement without a strnlen by always allocating size+1 bytes,
which is why the standard can put strnlen_s in an annex.

--dho

[1]: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2454.pdf

On Fri, Dec 18, 2020 at 2:01 AM Sigrid Solveig Haflínudóttir
<ftrvxmtrx@gmail.com> wrote:
>
> > Because the terminating '\0' may not be here.
>
> Why wouldn't it be there?  Isn't strndup supposed to work on strings?
> Is there any (trustworthy standard-like) source that mentions
> non-terminated strings as something that should work with strndup?
>