Re: strverscmp is buggy in newlib 4.4.0 (was Cygwin 3.4.6)

On 2024-01-02 10:36, Brian Inglis via Cygwin wrote:
> On 2024-01-02 03:23, Bruno Haible via Cygwin wrote:
>> Here's a test case of strverscmp, from Dmitry Bogatov [1]
>> #include <string.h>
>> int main ()
>> {
>>    return strverscmp ("UNKNOWN", "2.2.0") <= 0;
>> }
>> It succeeds on glibc and musl libc 1.2.4, but fails on musl libc 1.2.3
>> and Cygwin 2.9.0 and 3.4.6.
>> The cause is apparently that Cygwin's strverscmp implementation was
>> borrowed from musl libc (Cygwin commit 59e09b6419cdf400be3c73b61ac9c22560dc397e)
>> at a time when musl libc's implementation was buggy. In musl libc, it is
>> meanwhile fixed, through
>> https://git.musl-libc.org/cgit/musl/commit/src/string/strverscmp.c?id=b50eb8c36c20f967bd0ed70c0b0db38a450886ba
>> [1] https://lists.gnu.org/archive/html/bug-gnulib/2024-01/msg00002.html
> 
> Issue is in newlib (x-post and followups set):
> https://sourceware.org/cgit/newlib-cygwin/tree/newlib/libc/string/strverscmp.c
> 
> patch with git log msg:
> https://git.musl-libc.org/cgit/musl/patch/src/string/strverscmp.c?id=b50eb8c36c20f967bd0ed70c0b0db38a450886ba

This musl patch applies cleanly to newlib (patch offsets) and is attached.

> Gnulib discussion:
> https://lists.gnu.org/archive/html/bug-gnulib/2024-01/msg00002.html
> 
> Musl thread starts:
> https://www.openwall.com/lists/musl/2022/11/06/1
> 
> bare patch without git log msg attached to:
> https://www.openwall.com/lists/musl/2022/11/08/1
-- 
Take care. Thanks, Brian Inglis              Calgary, Alberta, Canada

La perfection est atteinte                   Perfection is achieved
non pas lorsqu'il n'y a plus rien à ajouter  not when there is no more to add
mais lorsqu'il n'y a plus rien à retirer     but when there is no more to cut
                                 -- Antoine de Saint-Exupéry

From b50eb8c36c20f967bd0ed70c0b0db38a450886ba Mon Sep 17 00:00:00 2001
From: Rich Felker <[email protected]>
Date: Mon, 7 Nov 2022 22:17:55 -0500
Subject: fix strverscmp comparison of digit sequence with non-digits

the rule that longest digit sequence not beginning with a zero is
greater only applies when both sequences being compared are
non-degenerate. this is spelled out explicitly in the man page, which
may be deemed authoritative for this nonstandard function: "If one or
both of these is empty, then return what strcmp(3) would have
returned..."

we were wrongly treating any sequence of digits not beginning with a
zero as greater than a non-digit in the other string.
---
 src/string/strverscmp.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

(limited to 'src/string/strverscmp.c')

diff --git a/src/string/strverscmp.c b/src/string/strverscmp.c
index 4daf276d..16c1da22 100644
--- a/src/string/strverscmp.c
+++ b/src/string/strverscmp.c
@@ -18,9 +18,9 @@ int strverscmp(const char *l0, const char *r0)
 		else if (c!='0') z=0;
 	}
 
-	if (l[dp]!='0' && r[dp]!='0') {
-		/* If we're not looking at a digit sequence that began
-		 * with a zero, longest digit string is greater. */
+	if (l[dp]-'1'<9U && r[dp]-'1'<9U) {
+		/* If we're looking at non-degenerate digit sequences starting
+		 * with nonzero digits, longest digit string is greater. */
 		for (j=i; isdigit(l[j]); j++)
 			if (!isdigit(r[j])) return 1;
 		if (isdigit(r[j])) return -1;
-- 
cgit v1.2.1