See Issue #9933
Some work started in PR #5872, but I can’t understand what’s going on in there…
It fixed the problem I was having (compiling in OSX 10.10.5) if there were spaces anywhere in the absolute path of the source code. Successfully compiled and passed all rpc-tests in paths with and without spaces.
GNU Make with spaces in filenames seems to be a general issue.
I will continue to bang on this and see if I break anything else.
0FileNotFoundError: [Errno 2] No such file or directory: '/home/travis/build/bitcoin/bitcoin/build/bitcoin-i686-pc-linux-gnu/qa/rpc-tests/txn_doublespend.py --mineblock'
It’s trying to execute the script + parameter as one executable name.
subprocess.Popen() likes that better on those builds.
205@@ -205,7 +206,7 @@ def main():
206 if args.help:
207 # Print help for test_runner.py, then print help of the first script and exit.
208 parser.print_help()
209- subprocess.check_call((config["environment"]["SRCDIR"] + '/test/functional/' + test_list[0]).split() + ['-h'])
210+ subprocess.check_call([config["environment"]["SRCDIR"] + '/qa/rpc-tests/' + test_list[0][0]] + ['-h'])
248@@ -248,6 +249,7 @@ def run_tests(test_list, src_dir, build_dir, exeext, jobs=1, enable_coverage=Fal
249 results = BOLD[1] + "%s | %s | %s\n\n" % ("TEST".ljust(max_len_name), "PASSED", "DURATION") + BOLD[0]
250 for _ in range(len(test_list)):
251 (name, stdout, stderr, passed, duration) = job_queue.get_next()
252+ name = " ".join(name)
name is a list not a string. So this turns ['testname', '--arg'] back into 'testname --arg' before it is displayed
Ah, of course.
Mind keeping them separate so that’s more obvious? e.g. rename to test and print test[0] + test[1:] ?
Sorry I’m late to this party. @theuni asked me to take a look at this.
I really don’t like the approach of changing the list of tests into a list of lists. It’s torturing the test_runner code for the sake of a couple of test cases which take arguments. It’s also borked the formatting of the printout report (look at how max_len_name gets set to see how).
I think it’s much better (and less intrusive) to keep the test list as a list of strings, and then just change the Popen() call to be something like:
0test_argv = t.split()
1self.jobs.append((t,
2 time.time(),
3 subprocess.Popen([self.tests_dir + test_argv[0]] + test_argv[1:] + self.flags + port_seed,
4 universal_newlines=True,
5 stdout=log_stdout,
6 stderr=log_stderr),
7 log_stdout,
8 log_stderr))
This is very similar to what Popen() was calling before, it’s just that split() is only being called on the test name string, rather than the complete path. I think that fixes your original problem of test cases not working when there’s a space in the path.
Can you test that and see if it works for you?
EDIT: you’ll also need to do the same thing in the subprocess.check_call() call where it’s printing the help text, since that has the same bug (it’s calling split() on the filename with path instead of just the test string)
I like https://github.com/bitcoin/bitcoin/pull/9946/commits/16008c4d8dfd1539cefd98feb45d83e6389efd27 much better. My comment would be that rather than what you’re doing:
instead you should:
That seems reasonable. You’re splitting the test string into arguments based on whitespace (which is what a shell does anyway!)
Glad it works. I’ll happily review this once you’ve pushed the new commit.
Regarding the timeouts: the first thing I’d check is whether there are any bitcoin processes running. Sometimes the framework fails to clean up old processes if the test fails, which can lead to issues next time you run a test. If that’s not the issue, I’ve beefed up the test logs recently. You should be able to run combine_logs.py <test directory> -c | less -r to get a combined view of what happened in the test.
And if that doesn’t help, please open a new issue with the logs and @ me and I’ll take a look.