linux: add missing SOL_* constants #228
Conversation
|
(rust_highfive has picked a reviewer for you, use r? to override) |
|
Looks like CI says some constants aren't defined? Perhaps a missing header to include? |
|
Yes, working on fixing up the tests. Bionic/kernel define several that aren't defined for musl/glibc. I think it is reasonable to define them all for Linux platforms, so I'll push an amended commit once I think I have things in place. |
|
Ok, thanks! Feel free to ping the PR once tests are passing and I will r+ |
posborne
force-pushed
the
socket-constants-sol
branch
2 times, most recently
from
6fc5783 to
f679d3d
Compare
SOL_SOCKET appears to be the only constant that is not the same across architectures and it was already present in libc. The constants provided match bionic and the Linux kernel but are a supseret of those provided by musl and glibc. They are just constants, so it should be fine (and avoids upstream from having to define/maintain them). Signed-off-by: Paul Osborne <osbpau@gmail.com>
posborne
force-pushed
the
socket-constants-sol
branch
from
f679d3d to
3a01937
Compare
|
Ok @alexcrichton, should be ready for review. Appveyor is a bit backed up, but things are passing on travis. |
| "SOL_LLC" | | ||
| "SOL_DCCP" | | ||
| "SOL_NETLINK" | | ||
| "SOL_TIPC" if linux => true, |
There was a problem hiding this comment.
Hm I'm a little wary of including constants that aren't in glibc or musl yet. Perhaps these could be android-only for now? Or should we update glibc to start verifying them?
There was a problem hiding this comment.
Working with glibc to verify these seems like it might be a reasonable approach although they probably move forward at a snails pace for fear of breaking existing code that has already defined constants missing from its headers (I found several major projects defining SOL_NETLINK themselves, some without an #ifndef check). That is, I think the lack of definition of these constants might have less to do with stability and more to do with the inability to have nice things when working in C.
Another angle would be to verify some definitions against the kernel headers. I don't see a good way of mixing glibc headers and kernel headers together, but having a separate set of tests which verify against kernel sources might have some legs.
There was a problem hiding this comment.
Hm ok, I'm largely just wary to include such a large set of constants in the ignore list. This'd mean that they're never validated, and I don't necessarily trust architectures like MIPS to have all the same values for all the constants. It's ok if they're not tested on MUSL, but they should be tested somewhere.
There was a problem hiding this comment.
Understood. I'll see what I can come up with. I would be fine with dropping all of these as well except SOL_NETLINK if it comes to that. They are tested in Android, so there is some validation going on (but we do not validate Android on non-ARM).
|
Closing due to inactivity, but feel free to resubmit with comments addressed! |
* sse2: _mm_stream_si128,si32,pd,si64 * sse2: _mm_stream_* tests * Disable assert_instr for _mm_stream_si64
SOL_SOCKET appears to be the only constant that is not the
same across architectures and it was already present in libc.
Signed-off-by: Paul Osborne osbpau@gmail.com