Use llvm-args instead of RUST_TEST_THREADS, %Nm instead of NEXTEST_TEST_THREADS

[taiki-e]

cargo-llvm-cov no longer sets the RUST_TEST_THREADS environment variable and uses llvm-args instead for workaround rust-lang/rust#91092.

TODO: skip passing llvm-args on newer nightly that includes rust-lang/rust#111469.

Closes #258
cc @Dushistov

[taiki-e]

Ah, we should also pass -fprofile-update=atomic to clang.

cargo-llvm-cov/src/main.rs

Line 278 in 752a20e

let clang_flags = " -fprofile-instr-generate -fcoverage-mapping";

[Dushistov]

It works and works correctly as far as I tested. But I don't see any time difference.
I tried functional and unit tests in one crate, like #258 .
Also I tried run tests for workspace with several crates there is no difference in work time for cargo llvm-cov nextest.

[taiki-e]

As for NEXTEST_TEST_THREADS, I have replaced it with the way described in #258 (comment). I have confirmed that #258 is fixed, but have not yet been able to test whether the original problem is completely fixed.

$ cargo llvm-cov nextest
   Compiling a v0.1.0 (/Users/taiki/projects/tmp/a)
    Finished test [unoptimized + debuginfo] target(s) in 0.14s
    Starting 2 tests across 2 binaries
        PASS [   8.015s] a tests::test_8
        PASS [  30.015s] a::functional_tests test_30
------------
     Summary [  30.018s] 2 tests run: 2 passed, 0 skipped

[Dushistov]

As for NEXTEST_TEST_THREADS, I have replaced it with the way described i

As result, on my real code base, I got such results:
before this PR: 556.727s
after this PR. : 182.539s

Great!

The coverage numbers are little different:

-Totals       67.03% (1545/2305)   79.55% (18853/23699)   67.27% (8344/12403) - (0/0)
+Totals       67.03% (1545/2305)   79.51% (18842/23699)   67.20% (8335/12403) - (0/0)

but I use https://github.com/BurntSushi/quickcheck that uses random generator,
so it is expected.

So, I finally can add all crates, not only important ones, in my workspace into coverage calculation for every CI run.

[svenstaro]

Could this be merged? From what I can tell, this is good to go, no?

[taiki-e]

Published in v0.5.20. Thanks @Dushistov for testing this!