Adding QNX support to googletest

[Yuryu, 2011-06-05 00:08:58]

Adding QNX support to googletest.

As of QNX 6.5.0, it does not support fork under multithread environment, so I
needed to use spawn(QNX specific system call) instead of fork.

[Vlad, 2011-06-11 19:48:56]

http://codereview.appspot.com/4517138/diff/1/src/gtest-death-test.cc
File src/gtest-death-test.cc (right):

http://codereview.appspot.com/4517138/diff/1/src/gtest-death-test.cc#newcode979
src/gtest-death-test.cc:979: const bool use_fork =
GTEST_FLAG(death_test_use_fork);
Thes death_test_use_fork flag only controls whether to use fork() over clone()
on Linux. The choice you make here is between forking or spawning a child, i.e.
between the fast and the thread-safe ways, is controlled by the death_test_style
flag. Consequently, you should add the QNX code here unconditionally, e.g.,

#if GTEST_OS_QNX
 ...qnx code...
#else
# if GTEST_HAS_CLONE
  ...
# endif
  if (use_fork && (child_pid = fork()) == 0) {
    ExecDeathTestChildMain(&args);
    _exit(0);
  }
#endif  // GTEST_OS_QNX

  GTEST_DEATH_TEST_CHECK_(child_pid != -1);
  return child_pid;

The NoExecDeathTest class should handle the fork case without problems, AFAICT -
but check just in case.

[Yuryu, 2011-06-12 06:23:23]

Thanks for reviewing.
I changed the code structure according to your comment.

[Vlad, 2011-06-12 09:20:40]

http://codereview.appspot.com/4517138/diff/4001/src/gtest-death-test.cc
File src/gtest-death-test.cc (right):

http://codereview.appspot.com/4517138/diff/4001/src/gtest-death-test.cc#newco...
src/gtest-death-test.cc:1014: #  endif  // GTEST_OS_QNX
Can you please move the #endif beyond the following if statement and remove the
use_fork variable from the QNX case? That makes it clear to someone reading the
code that that if statement is never used under QNX.

[Yuryu, 2011-06-12 10:57:09]

Excluded code from complication for QNX by if/endif according to reviewer's
comment.

http://codereview.appspot.com/4517138/diff/4001/src/gtest-death-test.cc
File src/gtest-death-test.cc (right):

http://codereview.appspot.com/4517138/diff/4001/src/gtest-death-test.cc#newco...
src/gtest-death-test.cc:1014: #  endif  // GTEST_OS_QNX
On 2011/06/12 09:20:40, Vlad wrote:
> Can you please move the #endif beyond the following if statement and remove
the
> use_fork variable from the QNX case? That makes it clear to someone reading
the
> code that that if statement is never used under QNX.

Done.
Also excluded ExecDeathTestChildMain from compilation because it is also not
used under QNX.

[Vlad, 2011-06-13 00:22:42]

It looks good now. Have you signed the CLA. Once I can verify that you did, I
will merge it into the codebase and commit to svn. It may take awhile, though -
please be patient. I will let you know when it lands in the head.

Regards,
Vlad

[Yuryu, 2011-06-13 04:38:50]

Thank you for reviewing again.
Yes, I signed the CLA as Haruka Iwao.

[Vlad, 2011-06-17 18:32:55]

http://codereview.appspot.com/4517138/diff/9001/src/gtest-death-test.cc
File src/gtest-death-test.cc (right):

http://codereview.appspot.com/4517138/diff/9001/src/gtest-death-test.cc#newco...
src/gtest-death-test.cc:965: const int cwd_fd = open(".", O_RDONLY);
A question -

Can the save/restore of the current directory be achieved by using more
cross-platform getcwd() and chdir() instead of platform-specific functions? That
may allow reusing this code on other platforms that do not support fork (e.g.,
Windows).

[Yuryu, 2011-06-20 02:21:30]

According to POSIX, chdir() fails on some platforms when the path is longer than
PATH_MAX.
In this case the program should pass each path section separately to chdir().
Another reason I used fchdir() is I can save memory when a testing program
resides in a very deep directory(and memory pressure is high on embedded
systems).
Both open()+fchdir() and getcwd()+chdir() are POSIX functions, so I don't think
this affects portability.

See http://pubs.opengroup.org/onlinepubs/9699919799/functions/getcwd.html

I don't know whether fchdir() or similar functions exist on Windows.