Rongsong Shen

I have checked the test fail cases. The fail should not be introduced by the code in pull request.

> Thanks for the PR. Please squash the second commit to the first, we don't need to format change separately. > > Also I'm aware of the limitation on the...

> Hmm.. this change makes almost every tests failing. Can you give more detail information?

> It fails with NZ meaning 'non-zero return value'. > > ``` > $ git log -2 --oneline --no-decorate > 3c9dad8a mcount: Do not keep FD for message channel >...

And something error with '-j8'(It seems that confliction between test cases) which should not be related to this PR. ```sh uftrace: /data/users/rshen/works/uftrace/cmds/recv.c:49:server_socket ERROR: socket bind failed (port: 45721): Address already...

test in my environment: ```sh -bash-4.2$ make runtest TESTARG='-vpO2 001' TEST test_run Start 1 tests without worker pool Compiler gc Runtime test case pg ------------------------: O2 build command: gcc -o...

Also the test result of the newest master: ```sh Start 1 tests without worker pool Compiler gc Runtime test case pg ------------------------: O2 build command: gcc -o t-abc -fno-inline -fno-builtin...

running strace with command: ```sh strace -f -o /tmp/log -v -s 1024 /data/users/rshen/works/uftrace/uftrace live --no-pager --no-event --libmcount-path=/data/users/rshen/works/uftrace -N memcpy t-abc ``` and check output of uftrace data file operations: ```sh...

@namhyung can you try to run strace -f on "/home/namhyung/project/uftrace/uftrace live --no-pager --no-event --libmcount-path=/home/namhyung/project/uftrace -N memcpy t-abc" ? Rongsong

> I'm not sure if it behaves differently depending on kernel version but it seems `pipe_poll()` in the kernel would set `(E)POLLHUP` bit in the `pollfd.revents` when there's no writers...