RFR: 8345016: [ASAN] java.c false positive ‘%s’ directive argument is null [-Werror=format-truncation=] [v2]

[Jaikiran Pai]

On Wed, 4 Dec 2024 16:04:01 GMT, SendaoYan <syan at openjdk.org> wrote:

>> Hi all,
>> The file src/java.base/share/native/libjli/java.c generate  false positive compile warning by gcc14 with gcc options `-fsanitize=undefined -O2`, and make jdk build error when configure with option `--enable-asan --enable-ubsan`. So I think it's necessory to disable the false positive compile warning to make jdk build normally with configure option `--enable-asan --enable-ubsan`.
>> 
>> This PR use gcc `_Pragma` to disbale the `-Wformat-truncation` compile warning, to make the disable code area as small as possible, and make the change looks ugly. There is another solution to disable the compile warning seen as below, but it will disable the compile warning of java.c. So I use the first solution.
>> 
>> 
>> 
>> diff --git a/make/modules/java.base/lib/CoreLibraries.gmk b/make/modules/java.base/lib/CoreLibraries.gmk
>> index 61ac495968a..5bc83cf0978 100644
>> --- a/make/modules/java.base/lib/CoreLibraries.gmk
>> +++ b/make/modules/java.base/lib/CoreLibraries.gmk
>> @@ -178,6 +178,7 @@ $(eval $(call SetupJdkLibrary, BUILD_LIBJLI, \
>>      OPTIMIZATION := HIGH, \
>>      CFLAGS := $(LIBJLI_CFLAGS) $(LIBZ_CFLAGS), \
>>      DISABLED_WARNINGS_gcc := unused-function unused-variable, \
>> +    DISABLED_WARNINGS_gcc_java.c := format-truncation, \
>>      DISABLED_WARNINGS_clang := deprecated-non-prototype format-nonliteral \
>>          unused-function, \
>>      DISABLED_WARNINGS_clang_java_md_macosx.m := unused-variable, \
>
> SendaoYan has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Fix compile warning C4068: unknown pragma 'GCC' on windows

The part that I don't understand is why the `s` is being considered as having a `NULL` value for the gcc compiler to generate that warning for the `snprintf` call:


if (s == NULL)
    return;
s = JLI_WildcardExpandClasspath(s);
if (sizeof(format) - 2 + JLI_StrLen(s) < JLI_StrLen(s))
    // s is became corrupted after expanding wildcards
    return;
size_t defSize = sizeof(format)
                   - 2 /* strlen("%s") */
                   + JLI_StrLen(s);
def = JLI_MemAlloc(defSize);
snprintf(def, defSize, format, s);
``` 
It looks like the compiler is considering `s = JLI_WildcardExpandClasspath(s);` to be returning `NULL` (if I comment that line out) the warning is gone. The other part to this is, I can't get the warning to be generated with gcc 13.x. So I don't know if this is a bug in gcc 14.x or some kind of improvement in detecting some genuine issue.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/22546#issuecomment-2545418968