On 29/11/2019 06:29, Thomas Stüfe wrote:
Hi Matthias,
I am not certain the callers are prepared to handle NULL.
This is used in a chain of TRANSFORM macro calls which AFAICS do not
handle NULL; e.g. , at 872, we pass the returned pointer to
convertUft8ToPlatformString which passes it on (on Windows) to
MultiByteToWideChar, which does not handle NULL input.
So I wonder whether a clear error message with an exit would be better
in this case. Otherwise we may get a crash just some instructions later.
Right, this needs a lot more analysis to see if it's even possible to
continue. The main usage is VM startup where the -javaagent option
specifies agents that have the Boot-Class-Path attribute. In that case
it would not be unreasonable to abort the process, it's unlikely to get
much startup in the startup if memory is exhausted. The other possible
context is where a tool agent is loaded into a running VM, in which case
have the attach thread return with a pending exception might be okay
(although the VM is likely to shutdown anyway as the memory exhaustion
will be detected/handled elsewhere).
-Alan
