Testing membership in V4 default route

[mirage]

Testing membership in V4 default route

[hannesm]

Thanks for the report, from the documentation in the Int32 module, I read "The result is unspecified if [y < 0] or [y >= 32]." (where shift_right_logical x y is considered) --> the calling code (this ipaddr library) needs to treat y = 32 specially. Since you already found a reproduction test case, it would be nice to add it to the test suite in lib_test -- and at the same time look into the IPv6 Prefix module which may repeat a similar issue.

[verbosemode]

I also ran some manual tests for IPv6 and the behavior looks correct. mem calls mask, which in turn calls V4.mask with the prefix length as argument. The latter returns 0l for prefix length of zero and bellow.

    # Ipaddr.V6.Prefix.mem (Ipaddr.V6.of_string_exn "2001:db8:c0ff:ee::1") 
           (Ipaddr.V6.Prefix.of_string_exn "::/0");;
    - : bool = true

I'll see if we can adapt the procedure of the v6 code to v4 and open a PR with the tests as you've suggested.

[hannesm]

thanks for looking into this @verbosemode -- looks like the V6 code is fine and doesn't need changes.
It would be great to have (a) failing tests for Ipv4.Prefix.mem (b) a fix for the issue (by handling 32 in a special way). If you need a hand (or further discussion / pointers), please let me know. We can as well look further into fixing this via IRC/XMPP.

[verbosemode]

The undefined behavior of shift_right_logical seems to manifest only in the toplevel or when explicitly compiling to bytecode.

I've created PR #98 to fix the issue and added some tests for V4.Prefix.mem. The existing tests used Prefix.mem, which in turn calls V6.Prefix.mem and might not hit code in V4.Prefix.mem.