Eliminate shared this from std/socket.d

[andralex]

This ends up simplifying Posix a lot. It could be simplified even further.

[andralex]

cc @CyberShadow @WalterBright @braddr @wilzbach @jmdavis @Imagix

[CyberShadow]

I think we can now get rid of getnameinfo being conditional by now. It was starting to lose sense when it was introduced in 2011, and today we probably don't support anything that old because of other reasons.

[CyberShadow]

BTW, I'm skeptical about replacing static ~this with atexit.

• Is atexit thread-safe? man says it's thread-safe but what about TLS variables? Will the function specified to atexit be called in the same thread as the code calling atexit?
• Do the semantics match up correctly? Such as thread/process distinctions, and unloading of DLLs?
• atexit seems to have a limit of 32 functions (ATEXIT_MAX).
• Not sure about the merits of replacing a D feature with a C feature.

[andralex]

I think we can now get rid of getnameinfo being conditional by now.

OK, I'll look into that.

I think we're good with atexit only as long as we only mess with other C libraries.

[andralex]

@CyberShadow just to make sure: on Posix getnameinfo, getaddrinfo, and freeaddrinfo can never be null weak symbols. Correct?

[CyberShadow]

Needs updates to documentation (e.g. lots of references to SocketFeatureException don't apply any more).

[andralex]

@CyberShadow done.

All - I notice that setKeepAlive throws depending on a statically known condition, which seems kinda silly. Should we just static assert(false) in that case? Then we can remove SocketFeatureException altogether.

[CyberShadow]

Nice work. Would be good to see the project tester's results for this one (so no auto-merge so far).

All - I notice that setKeepAlive throws depending on a statically known condition, which seems kinda silly.

Yeah, that's a bit of a Java-ism.

Should we just static assert(false) in that case?

It would mean making the function a template. Currently it's virtual due to virtual-by-default, and because of that Socket methods are all pretty sensitive to breakage (see issue 16514).

I guess there's also the small risk that there is code out there that already has setKeepAlive behind a regular if (not static).

[andralex]

Cool, so I guess this PR is good to go. @CyberShadow ?

[CyberShadow]

import core.stdc.stdlib : atexit; as per CircleCI

[andralex]

Urgh. Not a fan of this rule. I recall we introduced it to compensate for hijacking, which we since fixed.

[CyberShadow]

Still lots of errors in the autotester.

[dlang-bot]

Thanks for your pull request, @andralex!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

[andralex]

OK, pushed fixes to the autotester problems. Flying blindly here...

[andralex]

moar fixes

[andralex]

Hmmm, so now we have a different problem:

Error 42: Symbol Undefined _getnameinfo@28

It seems this is caused by switching from dynamic loading to statically linking. What library do we need to link with, and what's the cmdline syntax? Thx!

[CyberShadow]

You'll need to update the import libraries... :/

There are some definition files in druntime/def, but none for winsock. The best way may be to have the DMC import libraries updated. @WalterBright?

[WalterBright]

What library do we need to link with, and what's the cmdline syntax?

Use grep, the all-purpose tool:

grep getnameinfo \dm\lib\*.lib

Unfortunately, no hits. Trying:

grep getnameinfo \dm\bin\*.dll

No hits there, either. I don't know where it is supposed to be; it isn't in the DMC distribution.

[CyberShadow]

I don't know where it is supposed to be; it isn't in the DMC distribution.

You need to add getnameinfo and getaddrinfo (possibly others?) to ws2_32.lib.

[WalterBright]

You need to add getnameinfo and getaddrinfo (possibly others?) to ws2_32.lib.

The dm\bin\coffimplib.exe program can be used to convert 32 bit MS-Coff files to OMF.

[andralex]

Reverted the pull that causes problems on Windows.

[andralex]

This should be good to go subject to the autotester passing.

[andralex]

I'm throwing away this entire pile of dung and start anew.

[andralex]

Redone in #5813