Fix handling of ignore in wait() and wait_with_all()

[delan]

the logic in CmdChildren and FunChildren is currently a bit organic and inconsistent. as a result, the following test cases (also added in this patch) return the wrong results:

assertion failed (false != true): spawn!(ignore false).unwrap().wait().is_ok() when pipefail is on
assertion failed (false != true): spawn!(ignore false).unwrap().wait().is_ok() when pipefail is off
assertion failed (false != true): spawn!(ignore true | false).unwrap().wait().is_ok() when pipefail is on
assertion failed (false != true): spawn!(ignore true | false).unwrap().wait().is_ok() when pipefail is off
assertion failed (false != true): spawn!(ignore false | true).unwrap().wait().is_ok() when pipefail is on
assertion failed (false != true): spawn!(ignore false | false).unwrap().wait().is_ok() when pipefail is on
assertion failed (false != true): spawn!(ignore false | false).unwrap().wait().is_ok() when pipefail is off
assertion failed (false != true): spawn_with_output!(ignore false).unwrap().wait_with_all().0.is_ok() when pipefail is on
assertion failed (false != true): spawn_with_output!(ignore false).unwrap().wait_with_all().0.is_ok() when pipefail is off
assertion failed (false != true): spawn_with_output!(ignore true | false).unwrap().wait_with_all().0.is_ok() when pipefail is on
assertion failed (false != true): spawn_with_output!(ignore true | false).unwrap().wait_with_all().0.is_ok() when pipefail is off
assertion failed (false != true): spawn_with_output!(ignore false | false).unwrap().wait_with_all().0.is_ok() when pipefail is on
assertion failed (false != true): spawn_with_output!(ignore false | false).unwrap().wait_with_all().0.is_ok() when pipefail is off

this patch fixes wait() and wait_with_all() to correctly handle ignore. it also adds a comprehensive test suite that verifies that all of the ways you can get a CmdResult or FunResult (“entry points”) correctly handle the ignore command and pipefail mode. note that:

• wait_with_pipe() suffers from an unrelated and more complicated bug, so it’s not yet included in the test suite
• set_pipefail() is not thread safe, so the new tests have to live in test_pipe() for now, otherwise both will often fail

[delan]

to see the tests fail with the old code, check out this branch, then undo the changes to src/child.rs, e.g.

$ git remote add delan [email protected]:delan/rust_cmd_lib.git
$ git switch -c fix-ignore delan/fix-ignore
$ git restore -Ws @~ src/child.rs
$ cargo test

[tao-guo]

this is precise and thanks for the clean up.

[tao-guo]

Have you thought about using sealed_test to make the tests work ? Or we can make the CI to run tests sequentially

[delan]

we could do that, but i think it’s possible to fix the thread unsafety, and doing so would make the API more reliable. i’ll do that in a separate patch.