[PATCH 1/2] perf test python: Robustify the 'perf test python' test case

[Arnaldo Carvalho de Melo <acme@kernel.org>]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

While working on a patch to not build the 'perf test' entry that tests
the python binding when NO_LIBPYTHON=1 is used when building perf,
meaning that the python binding will not be built, thus no need to test
it, I noticed this inconsistency:

  $ perf test 17
   17: 'import perf' in python                                         : Ok
  $ perf test -F 17
   17: 'import perf' in python                                         : Ok
  $
  $ perf check feature libpython
               libpython: [ OFF ]  # HAVE_LIBPYTHON_SUPPORT
  $ ldd ~/bin/perf | grep python
  $

Even without any python binding or support for loading it present in
perf, it says that testing that feature somehow "passes":

  $ strace -s1024 -f -e execve perf test 17
  execve("/home/acme/bin/perf", ["perf", "test", "17"], 0x7ffe99ae5d50 /* 38 vars */) = 0
  strace: Process 519319 attached
   17: 'import perf' in python                                         : Running (1 active)
  strace: Process 519320 attached
  [pid 519320] execve("/bin/sh", ["sh", "-c", "--", "echo \"import sys ; sys.path.insert(0, '/tmp/build/perf-tools-next/python'); import perf\" |  2> /dev/null"], 0x377ba9a0 /* 40 vars */) = 0
  strace: Process 519321 attached
  strace: Process 519322 attached
  [pid 519321] +++ exited with 0 +++
  [pid 519322] +++ exited with 0 +++
  [pid 519320] --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=519321, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
  [pid 519320] +++ exited with 0 +++
  [pid 519319] --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=519320, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
  [pid 519319] +++ exited with 0 +++
   17: 'import perf' in python                                         : Ok
  +++ exited with 0 +++
  $

It doesn't matter if we fork a new perf process to run just that test
entry or if we don't:

  $ perf test -h -F

   Usage: perf test [<options>] [{list <test-name-fragment>|[<test-name-fragments>|<test-numbers>]}]

      -F, --dont-fork       Do not fork for testcase
  $

  $ strace -s1024 -f -e execve perf test -F 17
  execve("/home/acme/bin/perf", ["perf", "test", "-F", "17"], 0x7ffda8fafed8 /* 38 vars */) = 0
  strace: Process 519336 attached
  [pid 519336] execve("/bin/sh", ["sh", "-c", "--", "echo \"import sys ; sys.path.insert(0, '/tmp/build/perf-tools-next/python'); import perf\" |  2> /dev/null"], 0x159d99a0 /* 40 vars */) = 0
  strace: Process 519337 attached
  strace: Process 519338 attached
  [pid 519337] +++ exited with 0 +++
  [pid 519338] +++ exited with 0 +++
  [pid 519336] --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=519337, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
  [pid 519336] +++ exited with 0 +++
  --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=519336, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
   17: 'import perf' in python                                         : Ok
  +++ exited with 0 +++
  $

The system() call (that execve) will return zero even with that echo
being piped into nothing:

  # sh -c -- echo \"import sys ; sys.path.insert(0, '/tmp/build/perf-tools-next/python'); import perf\" |  2> /dev/null
  -bash: syntax error near unexpected token `0,'
  # echo $?
  2
  # sh -c -- echo \"import sys ; sys.path.insert(0, '/tmp/build/perf-tools-next/python'); import perf\" |
  -bash: syntax error near unexpected token `0,'
  # echo $?
  2
  #

If we instead avoid the echo and use 'python -c' to pass that simple
python script just trying to load the non-existent perf binding we get
less processes and a more consistent result even in this pathological
case where PYTHON="":

  $ perf test 17
   17: 'import perf' in python                                         : FAILED!
  $ perf test -F 17
   17: 'import perf' in python                                         : FAILED!
  $
  $ perf test -vv 17
  Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc
   17: 'import perf' in python:
  --- start ---
  test child forked, pid 522859
  python usage test: " -c "import sys ; sys.path.insert(0, '/tmp/build/perf-tools-next/python'); import perf" "
  sh: line 1: -c: command not found
  ---- end(-1) ----
   17: 'import perf' in python                                         : FAILED!
  $

The next patch will sidestep all this by plain not building the python
binding test when the binding isn't built, i.e. with NO_LIBPYTHON=1.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/tests/python-use.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/tests/python-use.c b/tools/perf/tests/python-use.c
index 0ebc22ac8d5b47ed..b7325caad22bab10 100644
--- a/tools/perf/tests/python-use.c
+++ b/tools/perf/tests/python-use.c
@@ -14,8 +14,8 @@ static int test__python_use(struct test_suite *test __maybe_unused, int subtest
 	char *cmd;
 	int ret;
 
-	if (asprintf(&cmd, "echo \"import sys ; sys.path.insert(0, '%s'); import perf\" | %s %s",
-		     PYTHONPATH, PYTHON, verbose > 0 ? "" : "2> /dev/null") < 0)
+	if (asprintf(&cmd, "%s -c \"import sys ; sys.path.insert(0, '%s'); import perf\" %s",
+		     PYTHON, PYTHONPATH, verbose > 0 ? "" : "2> /dev/null") < 0)
 		return -1;
 
 	pr_debug("python usage test: \"%s\"\n", cmd);
-- 
2.47.0