You would think that memchr can be used to test whether a string is
longer than N without traversing it. For instance we can take a
gigabyte-long character string and efficiently test wheether it is
shorter than 10 characters:

memchr(gigastr, 0, 10) == 0

if a null is found within the first 10 bytes, then its length
is 10 or more.

But suppose a 7 byte string is passed (length 6).

That *object* is smaller than n; it does not have an "initial sequence
of n characters" for memchr to search.

ISO C doesn't say that bytes of the initial sequence which are
beyond are sought-after value shall not be accessed by memchr.

For instance, for shits and giggles, memchr could perform a
right-to-left scan, and report the most recently found, hence
leftmost, occurrence of the value.

Or it could assume it can load an 8 byte word from the start of the object
(even if unaligned), since that lies within 10. Yet that 8 could extend
into an unmapped page.

--
TXR Programming Language: http://nongnu.org/txr
Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
Mastodon: @Kazinator@mstdn.ca

Kaz Kylheku <864-117-4973@kylheku.com> writes:

> You would think that memchr can be used to test whether a string is
> longer than N without traversing it. For instance we can take a
> gigabyte-long character string and efficiently test wheether it is
> shorter than 10 characters:
>
> memchr(gigastr, 0, 10) == 0
>
> if a null is found within the first 10 bytes, then its length
> is 10 or more.
>
> But suppose a 7 byte string is passed (length 6).
>
> That *object* is smaller than n; it does not have an "initial sequence
> of n characters" for memchr to search.
>
> ISO C doesn't say that bytes of the initial sequence which are
> beyond are sought-after value shall not be accessed by memchr. [...]

It does, and has for more than 10 years.

Kaz Kylheku <864-117-4973@kylheku.com> writes:

> You would think that memchr can be used to test whether a string is
> longer than N without traversing it. For instance we can take a
> gigabyte-long character string and efficiently test wheether it is
> shorter than 10 characters:
>
> memchr(gigastr, 0, 10) == 0
>
> if a null is found within the first 10 bytes, then its length
> is 10 or more.
>
> But suppose a 7 byte string is passed (length 6).
>
> That *object* is smaller than n; it does not have an "initial sequence
> of n characters" for memchr to search.
>
> ISO C doesn't say that bytes of the initial sequence which are
> beyond are sought-after value shall not be accessed by memchr.

Well, it does say that

"The implementation shall behave as if it reads the characters sequentially
and stops as soon as a matching character is found."

> For instance, for shits and giggles, memchr could perform a
> right-to-left scan, and report the most recently found, hence
> leftmost, occurrence of the value.
>
> Or it could assume it can load an 8 byte word from the start of the object
> (even if unaligned), since that lies within 10. Yet that 8 could extend
> into an unmapped page.

Only if the behaviour is consistent with the above quote, so anything
going wrong as a result of looking beyond the first occurrence is, I
think, ruled out.

--
Ben.

On 2023-09-03, Tim Rentsch <tr.17687@z991.linuxsc.com> wrote:
> Kaz Kylheku <864-117-4973@kylheku.com> writes:
>> ISO C doesn't say that bytes of the initial sequence which are
>> beyond are sought-after value shall not be accessed by memchr. [...]
>
> It does, and has for more than 10 years.

Thanks, Tim, and also Ben.

I looked in the wrong tab of the PDF reader, where I have C99 open!

--
TXR Programming Language: http://nongnu.org/txr
Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
Mastodon: @Kazinator@mstdn.ca