Re: [PATCH] Fix printf(1) for integer larger than INTMAX_MAX

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

On 2020/10/26 3:14, Robert Elz wrote:
>      Date:        Sun, 25 Oct 2020 19:01:18 +0900
>      From:        Rin Okuyama <rokuyama.rk%gmail.com@localhost>
>      Message-ID:  <09f349b6-f483-e582-d5b4-71f6b747db56%gmail.com@localhost>
>
>    | I forgot to mention in the previous message: %[oux] also fail for numbers
>    | larger than 1<<63:
>
> Yes, so they do - that's a (relatively) recent regression (which
> probably means I caused it) and should certainly be fixed.
>
>    | This is because we convert unsigned integers by strtoimax(3). So, I've
>    | updated the patch:
>    |
>    | http://www.netbsd.org/~rin/printf_20201025.patch
>
> Apart from the change to getwidth() (from ulong to uintmax_t) which isn't
> really needed, this is for handling field width values, and we arbitrarily
> silently truncate those (to a megabyte) already - an unsigned long can
> easily cope with any value we're actually going to use, the rest of that
> looks fine to me.

Ah, I didn't know ulong is intentionally. Thanks, I withdraw this part.

> There are probably (ugly) ways to deal with the commented out tests (because
> we have no simple way to know what range of values printf supports, so can't
> tell at what point errors, and value truncation, will start happening).  For
> the purposes of ATF tests (rather than generic sh programming) there are
> probably a sequence of operations that can be performed, where the results
> are not used for pass/fail of the test, but for deciding what later tests to
> run, perhaps using something like the method used in the sh arithmetic test
> which has the same needs (but which has the advantage of being able to use
> arithmetic to assist finding the range of values it supports).   That is,
> the details would vary, but the general method should be able to be made to
> work.

Hmm, it seems like tautology if printf(1) itself is used for this test.
How about this autoconf like method?

----
cat > test.c << EOF
#include <sys/types.h>
int main(void) { __CTASSERT(sizeof(intmax_t) == 8); return 0; }
EOF
if cc -c test.c 2> /dev/null; then
	echo "64 bit"
else
	echo "probably 128 bit or larger"
fi
rm -f test.o
----

This is straightforward, but, yes, ugly too ;).

Thanks,
rin