On Mon, 21 Sep 2026 07:23:53 GMT, Matthias Baesken <[email protected]> wrote:
>> src/jdk.jdwp.agent/share/native/libjdwp/signature.c line 49:
>>
>>> 47: if (tagPtr == NULL) {
>>> 48: EXIT_ERROR(AGENT_ERROR_NULL_POINTER, "Invalid method
>>> signature");
>>> 49: }
>>
>> request->methodSignature is setup by the following:
>>
>> error = methodSignature(method, NULL, &request->methodSignature, NULL);
>>
>> methodSignature() does the following:
>>
>> error = JVMTI_FUNC_PTR(gdata->jvmti,GetMethodName)
>> (gdata->jvmti, method, &name, &signature, &generic_signature);
>>
>> And then:
>>
>> if ( psignature != NULL ) {
>> *psignature = signature;
>>
>> psignature is the &request->methodSignature argument. So this is how it gets
>> setup. I don't see how strchr(signature, SIGNATURE_END_ARGS) can ever fail.
>> It would mean a bug in the JVMTI or the debug agent, which we should be
>> catching with an assert in debug builds as we currently do, not with an
>> EXIT_ERROR that applies to all builds.
>
> Could there be something overwritten in the signature in error cases, that
> leads to 'bad'/broken method signatures?
That sill sounds like a bug that should be caught with an assert. I don't see
how GCC can make any assumptions about a string returned by GetMethodName(). It
has to have a SIGNATURE_END_ARGS in it. GCC can't know that, but that shouldn't
lead to it to thinking strchr can return NULL in this case.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/32929#discussion_r4063453925
