Add Socket::IPaddress.v4, .v6, .v4_mapped_v6
#13422
Conversation
straight-shoota
changed the title
Socket.v4, .v6, .v4_mapped_v6Socket::IPaddress.v4, .v6, .v4_mapped_v6
| # windows returns the former which, while correct, is not canonical | ||
| # TODO: implement also `#to_s` in Crystal | ||
| {Socket::IPAddress.new("::ffff:0:0", 0), Socket::IPAddress.new("::ffff:0.0.0.0", 0)}.should contain Socket::IPAddress.v4_mapped_v6(UInt8.static_array(0, 0, 0, 0), 0) |
There was a problem hiding this comment.
Minor detail, but this expectation is a bit confusing.
How about being more explicit about which result is expected?
| # windows returns the former which, while correct, is not canonical | |
| # TODO: implement also `#to_s` in Crystal | |
| {Socket::IPAddress.new("::ffff:0:0", 0), Socket::IPAddress.new("::ffff:0.0.0.0", 0)}.should contain Socket::IPAddress.v4_mapped_v6(UInt8.static_array(0, 0, 0, 0), 0) | |
| # windows returns the former which, while correct, is not canonical | |
| # TODO: implement also `#to_s` in Crystal | |
| Socket::IPAddress.v4_mapped_v6(UInt8.static_array(0, 0, 0, 0), 0).should eq Socket::IPAddress.new({{ flag?(:win32) }} ? "::ffff:0:0" : ::ffff:0.0.0.0", 0) |
I'm happy either way.
There was a problem hiding this comment.
Like the previous Socket.ip? spec, I don't have proof that Windows actually returns the non-canonical IP address on all OS versions, so it is better to err on the side of caution
| sin_port: endian_swap(port), | ||
| sin_addr: LibC::InAddr.new(s_addr: endian_swap(addr_value)), | ||
| ) | ||
| new(pointerof(addr), sizeof(typeof(addr))) |
There was a problem hiding this comment.
@HertzDevil I just noticed this now during the review of #13463.
Taking a pointer to a variable on the stack is dangerous. It seems to be working here, but it's fragile.
It probably only works because .new and #initialize don't allocate enough variables to override the (relevant) parts on the stack.
I suppose a safer alternative could perhaps be a yielding variant of #initialize or some other way to directly assign to the memory of the Address struct.
Same applies for v6.
There was a problem hiding this comment.
There isn't a tail call here (I believe you have to opt in explicitly in the LLVM IR), so addr will be valid before new returns. The #initialize bodies exclusively dereference addr and do not attempt to extend its lifetime.
In any case I plan to make SockaddrIn or SockaddrIn6 part of the IPAddress itself so that #to_unsafe never allocates. That would allow addr to be passed by value here
There was a problem hiding this comment.
Okay, if LLVM keeps this safe, then it's fine.
Resolves #13382.