On Mon, 25 Nov 2024 07:08:56 GMT, David Holmes <dhol...@openjdk.org> wrote:
>> Hi all,
>> File `src/java.base/share/native/libjli/java.c` compile `error: control
>> reaches end of non-void function [-Werror=return-type]` with gcc options
>> `-fsanitize=address -O0`. The function `int JavaMain(void* _args)` in this
>> file will execute `return ret` in `LEAVE()` macro, but gcc with -O0 is not
>> smart enough to recognized that the function already has `return` statement
>> before at the end of function.
>> This PR add final return statement make gcc with options `-fsanitize=address
>> -O0` happy, to make jdk compile success with configure option
>> `--enable-asan` by slowdebug mode. The added `return ret` make no sence
>> because the `LEAVE()` macro make sure that function will execute `returun
>> ret` statement at java.c:348. So I think this change is no risk.
>>
>> Additional testing:
>>
>> - [x] jtreg tests(include tier1/2/3 etc.) with linux-x64 release build
>> - [x] jtreg tests(include tier1/2/3 etc.) with linux-x64 fastdebug build
>> - [x] jtreg tests(include tier1/2/3 etc.) with linux-aarch64 release build
>> - [x] jtreg tests(include tier1/2/3 etc.) with linux-aarch64 fastdebug
>> build
>
> src/java.base/share/native/libjli/java.c line 663:
>
>> 661: }
>> 662: LEAVE();
>> 663: return ret;
>
> A `do-while` loop is always executed at least once, that is why the `LEAVE
> macro is structured as it is:
>
> do { \
> if ((*vm)->DetachCurrentThread(vm) != JNI_OK) { \
> JLI_ReportErrorMessage(JVM_ERROR2); \
> ret = 1; \
> } \
> if (JNI_TRUE) { \
> (*vm)->DestroyJavaVM(vm); \
> return ret; \
> } \
> } while (JNI_FALSE)
>
> So this will always return and the sanitizer seems incorrect.
Yes, I was make a mistake prior, the do-while(false) loop will execute once.
But I think this PR still a bit usefull and no risk, the analyze seen as
[below](https://github.com/openjdk/jdk/pull/22355#issuecomment-2497997812).
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/22355#discussion_r1857554793
