Shouldn't memset structs containing std::string

[Ben Smith]

In the new C++ midas, several structs now have std::string members (e.g. callback_addr and RPC_CLIENT_CONNECTION in msystem.h). Unfortunately, there are still places that try to use memset to initialise/reset these structures. This is undefined behaviour, and leads to crashes in my system.

Example problematic calls are:

• mserver.cxx:306 - on my system spawned mserver processes segfault when trying to write to callback.host_name on L322, and we cannot run any remote frontends.
• midas.cxx:10987 - on my system clients segfault when later trying to destruct the client connection object.

There are at least half a dozen other instances just for the structs in msystem.h.

‌

I think we should define a common pattern for resetting such structures, and apply it consistently. A couple of immediately-obvious options are:

• Define a reset() function in each struct. We would have to remember to update it whenever the struct layout changes.
• Change them to nested structs, so the POD types are in an inner struct which can be memset. We would probably still need a reset() function to reset the strings to "", so I prefer the first option over this one.

There may be cleaner / simpler solutions I’m missing. Thoughts?

[Stefan Ritt]

I agree we should go with a reset() function for each structure. Or call it clear().

[Ben Smith]

This is now fixed in 9241f17 by implementing clear() for the relevant structs in msystem.h. There were also some places that used memcpy which I replaced with a call to the default copy-constructor.

We should be sure to check for memset/memcpy usage for any more structs that are converted to strings in future.