[Bug 210837] lang/perl5.22 (and related?): ext/re/re_exec.c has long long format matched up with long argument

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=210837

--- Comment #9 from Mark Millard <markmi at dsl-only.net> ---
(In reply to Mathieu Arnold from comment #7)

I quoted the code in question and then I said of it:

"the IVdf vs. (long)(rx_origin - strbeg) do not match types"

That is a (too short?) description of a way of being broken: it indicates that
the compiler's report was not a false positive.

In 32-bit long / 64-bit long long contexts the IVdf format uses more bytes (for
long long) then are put out on the stack for the value (long), picking out some
other bytes from memory as well. The details of the bad bytes use depends on
big-endian vs. little-endian (or other memory layouts for such values). For
little-endian if the extra bytes are zero bytes then the values would look
okay. Otherwise not.

The operational difficulty for live testing comes in knowing how to reach the
problem code in order to see the bad result. I'm not so sure that I'm likely to
hit the broken code in the implicit use of perl5 in my context.

(perl5 was built because of dependencies, not because I directly built it for
my own use.)

amd64 and powerpc64 would not be observably broken (long is also 64 bits).
powerpc would be broken (long is 32-bits and poewrpc is big-endian). armv6
would depends on if the extra bytes at the time happened to be zeros (long is
32 bits little-endian).

-- 
You are receiving this mail because:
You are the assignee for the bug.