illumos gate

Is there any reason to fix this bug? The realloc() call on that line is unlikely to fail because it extends the space by only one pointer. If it did fail, the subsequent malloc() would also fail. I can devise a change to prevent a core dump if realloc() failed, but is there any reason to do so?

Gary Mills wrote:

Is there any reason to fix this bug? The realloc() call on that line is unlikely to fail because it extends the space by only one pointer. If it did fail, the subsequent malloc() would also fail. I can devise a change to prevent a core dump if realloc() failed, but is there any reason to do so?

I see the main reason to fix this in a possibility to kill the bad habits with the realloc() usage. I believe we have a lot of similar pieces of the code through the gate, so it would be great to have some kind of project "review/fix all the realloc() callers" so the new code writers are not inspired by them.

As a start, I've reviewed all the illumos source, looking for instances where a realloc() failure is not handled by the code. I found 45 such files. These are listed in bug #5970 . One of them is the file in question here.

I can confirm that beadm does dump core in be_sort_list() whenever realloc() returns NULL. I tested this in Solaris 11.2, oi_151a9, and a recent build of illumos-gate. All of them dumped core.