[BUG] Changes to build fd on Illumos

[nealsid]

Checks

• I have read the troubleshooting section and still think this is a bug.

Describe the bug you encountered:

When the build system is running configure for jemalloc, an error is reported:

  configure: exit 1

  --- stderr
  Invalid configuration `x86_64-unknown-illumos': OS `illumos' not recognized
  configure: error: /bin/sh build-aux/config.sub x86_64-unknown-illumos failed

  thread 'main' panicked at /export/home/neal/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/tikv-jemalloc-sys-0.6.0+5.3.0-1-ge13ca993e8ccb9ba9847cc330696e02839f328f7/build.rs:\
388:9:

fd already disables jemalloc based on certain OS/architectures so Illumos can be added to this list. The GitHub project for jemalloc has been archived (https://github.com/jemalloc/jemalloc) so I don't know if it's possible to fix this upstream.

Describe what you expected to happen:

Nothing really, since Illumos isn't officially supported.

What version of fd are you using?

bac535ea

Which operating system / distribution are you on?

SunOS openindiana 5.11 illumos-23690bf77a i86pc i386 i86pc

[nealsid]

FYI, there are two test failures on Illumos I'm looking into (I know Illumos isn't a supported platform, just making a best effort to get it to work). The first one looks strange, it's as if -n is being printed to the console instead of processed as a command line switch to echo. If I run echo -n foo in a terminal on Illumos it behaves as expected (i.e. it suppresses printing a newline)

---- test_exec_multi stdout ----

thread 'test_exec_multi' panicked at tests/testenv/mod.rs:284:13:
`fd foo --exec echo -n {/}:  ; --exec echo {//}` did not produce the expected output.
Showing diff between expected and actual:
-C.Foo2: ./one/two
-a.foo: .
-b.foo: ./one
-c.foo: ./one/two
-d.foo: ./one/two/three
-directory_foo: ./one/two/three
+-n C.Foo2:
+-n a.foo:
+-n b.foo:
+-n c.foo:
+-n d.foo:
+-n directory_foo:
+.
+./one
+./one/two
+./one/two
+./one/two/three
+./one/two/three

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

---- test_invalid_cwd stdout ----

thread 'test_invalid_cwd' panicked at tests/tests.rs:2611:27:
called `Result::unwrap()` on an `Err` value: Os { code: 22, kind: InvalidInput, message: "Invalid argument" }


failures:
    test_exec_multi
    test_invalid_cwd

test result: FAILED. 98 passed; 2 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.66s

[nealsid]

Regarding the test failure in test_exec_multi, it turns out that when I run echo -n from the command line, it's using the shell built-in, but when fd (or find, which also reproduces the issue) calls echo, it's using /usr/bin/echo, which doesn't support -n:

neal@openindiana:/code/illumos-gate$ type -a echo
echo is a shell builtin
echo is /usr/bin/echo
echo is /usr/gnu/bin/echo
neal@openindiana:/code/illumos-gate$

Repro:

neal@openindiana:/code/illumos-gate$ /usr/bin/echo -n hello
-n hello
neal@openindiana:/code/illumos-gate$

[nealsid]

Regarding the second test failure in test_invalid_cwd, the way the test currently works is:

1. It creates a temporary directory
2. It creates a directory called foo off of the directory created in 1
3. Then, it sets the current directory to the foo subdirectory
4. Then, it deletes the foo subdirectory
5. Then the test runs fd with an argument indicating it should start in the directory created in 1.

On Illumos the test fails in step 4 because libc's rmdir() is documented as returning EINVAL when the directory to be removed is the current directory. This doesn't appear to be the case (AFAICT) on Linux.

I can create a similar test case from the command line with two terminals, by having one in the foo subdirectory and having the other terminal call rmdir on the foo folder, but doing this within one process does not appear possible. Even if we create multiple threads, chdir() works on a process-wide basis, meaning there's no way to have a thread maintain a cwd of the foo subdirectory while another thread is deleting it.

I guess the alternatives are to 1. Modify the test case to shell out to rmdir to delete foo in step 4 or 2. Disable the test on Illumos

[tmccombs]

I would probably be fine excluding that test for illumos. The behavior isn't very OS dependent, and we don't have CI for illumos anyway.