Date: Sun, 11 May 1997 02:20:02 -0700 (PDT) From: Bruce Evans <bde@zeta.org.au> To: freebsd-bugs Subject: Re: misc/3575: Eliminate spurious warning in /usr/src/lib/libc/stdlib/strtoq.c Message-ID: <199705110920.CAA19857@hub.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR misc/3575; it has been noted by GNATS.
From: Bruce Evans <bde@zeta.org.au>
To: FreeBSD-gnats-submit@freebsd.org, krw@tcn.net
Cc: Subject: Re: misc/3575: Eliminate spurious warning in /usr/src/lib/libc/stdlib/strtoq.c
Date: Sun, 11 May 1997 19:13:09 +1000
> During a make world the compilation of
>
> /usr/src/lib/libc/stdlib/strtoq.c
>
> produces a warning message
>
>
> /usr/src/lib/libc/stdlib/strtoq.c:108: warning: integer overflow in expression
This warning is a compiler bug (there is no overflow for unsigned types)
and isn't easy to fix. Note that there is no warning for corresponding
operation in strtol.c (the warning is for negating (u_quad_t)QUAD_MIN).
>--- /usr/src/lib/libc/stdlib/strtoq.c Sat May 10 11:55:38 1997
>+++ strtoq.c Sat May 10 14:10:02 1997
>@@ -105,7 +105,7 @@
> * overflow.
> */
> qbase = (unsigned)base;
>- cutoff = neg ? -(u_quad_t)QUAD_MIN : QUAD_MAX;
>+ cutoff = (u_quad_t)(neg ? QUAD_MIN : QUAD_MAX);
> cutlim = cutoff % qbase;
> cutoff /= qbase;
> for (acc = 0, any = 0;; c = *s++) {
This change would unimprove portabilty. E.g., for 64 bit 1's
complement, QUAD_MIN is -0x7fffffffffffffffLL (which is represented as
0x8000000000000000). Conversion to u_quad_t gives 0x8000000000000001ULL.
This must be negated (as in the original version) to give the correct
cutoff of 0x7fffffffffffffffULL.
For 1's complement the cutoff for the `neg' case can be expressed more
naturally as (u_quad_t)(-QUAD_MIN). This fails for 2's complement due
to overflow. I think the following expression works for 1's complement
and 2's complement and avoids the compiler bug:
cutoff = neg ? (u_quad_t)-(QUAD_MIN + 1) + 1 : QUAD_MAX;
This depends on QUAD_MIN not being too different from -QUAD_MAX. It can
be improved to:
cutoff = neg ? (u_quad_t)-(QUAD_MIN + QUAD_MAX) + QUAD_MAX : QUAD_MAX;
I think fixing the compiler would be easier :-).
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199705110920.CAA19857>