Avoid memcpy of invalid pointer

[MichaelChirico]

See r-lib/vctrs#1968 for more discussion

[jeroen]

Thanks. How is RAW_RO different from RAW ?

[MichaelChirico]

It returns the const pointer, which matches the signature of memcpy, so should be ever-so-slightly preferable (in general, no specific relation to the invalid pointers issue):

https://github.com/r-devel/r-svn/blob/1c8173fa21cb59afe3a95ff8c55d77aba2743481/src/include/Defn.h#L423-L430

https://www.geeksforgeeks.org/memcpy-in-cc/

[jeroen]

OK thanks. Btw did you actually experience a problem or is this mostly precaution?

[MichaelChirico]

Hm, mix of both :)

I haven't experienced any problem from this specific {curl} code, but our compiler does segfault when encountering the RAW() [and friends] output on R>=4.5.0, i.e. I've the risk of invalid/undefined behavior access is not purely theoretical.

So if it's possible to reach this code with sensible (&public API) inputs, that segfault can happen. I just can't judge without a much more careful reading if there are other checks set up elsewhere that would prevent a 0-length memcpy() in the first place.