refactor: conditional compile when glibc static-crt

[Itsusinn]

🤔 This is a ...

• Refactoring

[ibigbug]

why we need this change?

[Itsusinn]

1. We should use “real” SystemResolver as much as possible (before changes gnu with -crt-static use hickory for example)
2. Android dont support TokioResolver::tokio_from_system_conf but allow using “real” SystemResolver(libc's getaddrinfo)

[ibigbug]

i think the TokioResolver::tokio_from_system_conf was added for libc::getaddrinfo causing crash in certain cases, if we don't have that issue anymore we should look into remove the system TokioResolver some day

[Itsusinn]

i think the TokioResolver::tokio_from_system_conf was added for libc::getaddrinfo causing crash in certain cases, if we don't have that issue anymore we should look into remove the system TokioResolver some day

libc::getaddrinfo will fail under static gnu libc, it a glibc bug existing for decades.
That means only when we have static-crt of gnu targets, we have to keep system TokioResolver

[ibigbug]

ok fair engouth. thanks!

[ibigbug]

Maybe we should put a comment explaining the issue

[Itsusinn]

https://source.android.com/docs/core/ota/modular-system/dns-resolver?hl=zh-cn#dns-q

Android bonic libc's getaddrinfo is complicated, and broken on cross-test.

It works on real android device，tho.

[Itsusinn]

From cross-rs

[1] libc = bionic; Only works with native tests, that is, tests that do not depends on the Android Runtime. For i686 some tests may fails with the error assertion failed: signal(libc::SIGPIPE, libc::SIG_IGN) != libc::SIG_ERR, see cross-rs/cross#140 for more information.

I gonnna skip test for android target