Re: %.1s format with vasnprintf reads more than one byte from argument

[Eric Blake]

Eric Blake <ebb9 <at> byu.net> writes:

> git reset --hard e8e63
> ./gnulib-tool --with-tests --test vasprintf-posix
> 
> fails:
> 
> ../../gltests/test-vasprintf-posix.c:3624: assertion failed
> Abort trap 
> FAIL: test-vasprintf-posix
> 
> In the debugger, I see that vasprintf(&result,"%.*ls",2,L"ab") is setting 
> result to "a", not the expected "ab".
> 

On OpenBSD, this results in vasnprintf trying to invoke the native:

retcount = snprintf(buffer,12,"%.*ls%n",2,L"ab",&count);

which sets buffer to "a", and retcount and count to 1, rather than the 
expected "ab" and 2.  Indeed, further stepping through the OpenBSD source code 
shows that when precision is present, %.*ls uses memchr(arg,0,prec) to find 
whether a terminating NUL is present, forgetting that it should instead be 
looking for a terminating L'\0' and that the 8 bit '\0' found inside L'a' does 
not count as the end of the wide string.

Looks like we need to add to the printf-posix.m4 tests to filter out this bug 
in BSD %.*ls.

-- 
Eric Blake