Behavior of microros_allocate(0)?

[yushijinhun]

I found the behavior of microros_allocate() is inconsistent with RCL's default allocator implementation malloc(). When allocating zero size memory area, microros_allocate() returns NULL while malloc() (on Linux) returns a valid memory address.

I encounter this problem when trying to turn on RCL_COMMAND_LINE_ENABLED. I pass an empty commandline argument into rclc_support_init() (argc = 1, argv = {""}), and then ROS crashes at rcl_init because it tries to allocate 0 byte memory.

The workaround is quite simple: just allocate at least 1 byte in microros_allocate() if requested size is 0. However, in the manual of malloc(), it says:

NULL may also be returned by a successful call to malloc() with a size of zero, or by a successful call to calloc() with nmemb or size equal to zero.

I'm not sure whether it's a micro-ROS's bug or RCL's bug.

[pablogs9]

I guess that we comply with malloc() functionality. As far as we try to save as much memory as possible in the micro-ROS stack, maybe is better to fix the RCL code to handle the NULL return in the CLI parsing.

[yushijinhun]

I think the root of the problem is that RCL misses the terminating null byte when copying strings. I opened ros2/rcl#969 to solve this.