sys/net/sock_udp: Add missing const qualifier

[maribu]

Contribution description

The APIs sock_udp_get_local() and sock_udp_get_remote() are expected to be simple getters that do not modify the socket in any way. Hence, the should use cost sock_udp_t * rather than sock_udp_t * as first argument.

This fixes the issue.

Note: Since C callers will not be affected by this change, I will not add the API change label.

Testing procedure

The CI should be green.

Issues/PRs references

None

[riot-ci]

Murdock results

✔️ PASSED

a74c7df sys/net/sock_udp: Add missing const qualifier

Success	Failures	Total	Runtime
10197	0	10197	19m:55s

Artifacts

• Documentation preview

[miri64]

Are you sure that a change (even if small) to a major front-facing API can be classified as "minor impact"?

[miri64]

Also: why only change sock_udp? What about sock_tcp, sock_ip, and sock_dtls?

[maribu]

Are you sure that a change (even if small) to a major front-facing API can be classified as "minor impact"?

Yes, as C callers will not be affected. (Converting a foo * to const foo * works implicitly, as not writing to rw memory is not an issue. Dropping const would be an issue, as converting const foo * to foo * will not work implicitly, as writing to ro memory is an issue.)

Also: why only change sock_udp? What about sock_tcp, sock_ip, and sock_dtls?

I can take a look if similar issues are there as well. But let's fix one bug at a time.

[miri64]

Also: why only change sock_udp? What about sock_tcp, sock_ip, and sock_dtls?

I can take a look if similar issues are there as well. But let's fix one bug at a time.

To me its the same API, so IMHO it should be fixed in one go... But sock_udp seems to be the preferred API to work on anyways, so 🤷‍♀️.

[maribu]

For TCP making the sock_tcp_t * parameter const is not possible with the GNRC implementation, as sock_tcp_get_local() does indeed write to the socket (by locking a mutex allocated in the socket).

I'm closing this for now.

[maribu]

Revisiting this: I believe it is justified to have the TCP socket write-able: The connection sockets are not owned by the user of the API, but by the listen/queue socket that will freely reuse them when new connections get accepted. The mutex does only solve a data race, though: E.g. code may still assume that a socket is still belonging to connection A when it is calling sock_tcp_get_local_address(), but in fact the socket may already be reused for connection B. But there is only so much one can do with mutexes.

For UDP, the story is different. A user of the API will create the socket, and it can be assumed that the endpoint will not be changed while the caller is still owning the socket. I assume the same to be true for DTLS.

I suggest that I change the parameter to const therefore for UDP and DTLS only, and not for TCP.