On Wed, Aug 05, 2020 at 09:28:12PM +0300, Peter Pentchev wrote:
> On Wed, Aug 05, 2020 at 08:01:53PM +0200, Michał Mirosław wrote:
> > On Sun, Aug 02, 2020 at 05:34:40PM +0300, Peter Pentchev wrote:
> > > On Sun, Aug 02, 2020 at 02:02:22AM +0200, Michał Mirosław wrote:
> > [...]
> > > --- a/debian/tests/runtime
> > > +++ b/debian/tests/runtime
> > > @@ -432,6 +432,7 @@ MAIN:
> > >
> > > if (!defined $line) {
> > > $eof->send($got_version);
> > > + undef $f_out;
> > > } elsif (!$got_version) {
> > > if ($line =~ m{^
> > > stunnel \s+
> >
> > I believe $f_out will not be defined here, as it only gets set after
> > sub{} is created. Perl confirms this:
> >
> > $ TEST_STUNNEL=src/stunnel strace -f -o /tmp/log debian/tests/runtime
> > Global symbol "$f_out" requires explicit package name (did you forget to
> > declare "my $f_out"?) at debian/tests/runtime line 435.
> > Execution of debian/tests/runtime aborted due to compilation errors.
>
> Of course you're right. Sorry about that! That's what I get for writing
> a patch three minutes before I have to head out and never remembering to
> actually test it later :(
>
> How about the attached one?
[...]
> --- a/debian/tests/runtime
> +++ b/debian/tests/runtime
> @@ -424,7 +424,8 @@ MAIN:
>
> my ($got_version, $before_version) = (undef, '');
> my $eof = AnyEvent->condvar;
> - my $f_out = AnyEvent->io(
> + my $f_out;
> + $f_out = AnyEvent->io(
> fh => $s_in,
> poll => 'r',
> cb => sub {
> @@ -432,6 +433,7 @@ MAIN:
>
> if (!defined $line) {
> $eof->send($got_version);
> + undef $f_out;
> } elsif (!$got_version) {
> if ($line =~ m{^
> stunnel \s+
This stops the endless readings of EOF, but:
1. the FD gets leaked (shouldn't matter much, though)
2. the test hangs anyway
Using print-debugging, I see that it stops at wait_for_child line just
after printing the version. It seems that something is reaping the child
before the main thread has a chance to wait for it.
>From strace:
4285 +++ exited with 0 +++
4284 <... epoll_ctl resumed> ) = 0
4284 --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=4285,
si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
4284 write(4, "\1\0\0\0\0\0\0\0", 8) = 8
4284 rt_sigreturn({mask=[PIPE]}) = 0
4284 epoll_wait(3, [{EPOLLHUP, {u32=5, u64=4294967301}}, {EPOLLIN, {u32=4,
u64=4294967300}}], 64, 59743) = 2
4284 epoll_ctl(3, EPOLL_CTL_MOD, 5, {EPOLLIN, {u32=5, u64=4294967301}}) = 0
4284 read(4, "\1\0\0\0\0\0\0\0", 8) = 8
4284 wait4(-1, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}],
WNOHANG|WSTOPPED|WCONTINUED, NULL) = 4285
4284 wait4(-1, 0x7ffcd56d5784, WNOHANG|WSTOPPED|WCONTINUED, NULL) = -1 ECHILD
(No child processes)
4284 wait4(-1, 0x7ffcd56d5534, WNOHANG, NULL) = -1 ECHILD (No child processes)
This is before the 'wait_for_child' a few lines later.
4284 epoll_wait(3, [{EPOLLHUP, {u32=5, u64=4294967301}}], 64, 59743) = 1
4284 epoll_ctl(3, EPOLL_CTL_MOD, 5, {EPOLLIN, {u32=5, u64=4294967301}}) = 0
4284 read(5, "", 8192) = 0
4284 write(1, "Got stunnel version 5.56\n", 25) = 25
4284 write(2, "wait child 4285\n", 16) = 16
This is version printout plus my check.
4284 epoll_wait(3, [{EPOLLHUP, {u32=5, u64=4294967301}}], 64, 59743) = 1
4284 epoll_ctl(3, EPOLL_CTL_DEL, 5, 0x5631c0d44a40) = 0
This clears watcher for the pipe.
4284 epoll_wait(3, 0x5631c0d44a40, 64, 59743) = -1 EINTR (Interrupted system
call)
And this waits forever.
Best Regards
Michał Mirosław
