Avoid stringop-overflow errors #10293
Conversation
error: '__builtin_strncpy' writing 512 bytes into a region of size 1 overflows the destination Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
|
Tested by @knn-k - see #10278 (comment). |
|
#4486 is the original issue with trclog.c and AArch64. |
|
jenkins compile all jdk11 |
| * target for strcpy, etc.: we want to avoid warnings due to the | ||
| * apparent lack of space (we've made sure there's enough room). | ||
| */ | ||
| recordThreadName = offsetof(UtTraceRecord, threadName) + (char *)(void *)&trcBuf->record; |
There was a problem hiding this comment.
Do you need to cast twice here? I don't think (void *) is required.
There was a problem hiding this comment.
Perhaps not, but I think it's safer with the extra cast as it should make it clear to the compiler that there's no relationship between the pointer &trcBuf->record and the result of the casts.
Consider an analogy with Java: You'll get a compile error trying to cast a String to an Integer, but you can write (Integer)(Object)"oops" which compiles just fine (though you'll get a ClassCastException at runtime). The cast to Object is comparable to the cast to void * (and there's no exception in C).
There was a problem hiding this comment.
I expect this particular part of the change will be unnecessary once eclipse/omr#5444 is merged.
|
this is not a correct solution. see #10244 (comment), #10244 (comment), #10299 (comment) |
Please explain this assertion. |
|
Can one of the admins verify this patch? |
Rework code that sets thread name in trace buffer to avoid compiler errors, e.g.: