On 26 Jan 2021, Daniel Shahaf wrote:
>Yay! *gestures* Here's your old chair, right where you left it; I think you'll
>find the only difference from when you last occupied it is that your DHCP lease
>may have expired ;-)
:laugh+cry emoji: :)
>AFAICT:
>
>1. SVN_TEST_PYTHON is only used by the svneditor.bat, not by anything else in
>the test suite.
>
>2. When «make check» invokes foo_tests.py, it doesn't _execute_ it but
>_imports_ it, so the #! line doesn't even come into play at all. (See
>build/run_tests.py)
Thank you! Got it.
>The test suite should never run its own auxiliary Python scripts with
>«/usr/bin/env python»; rather, it should use sys.executable. There's precedent
>for this throughout the test suite.
>
>> So, one solution would be to just commit this change...
>>
>> Index: subversion/tests/cmdline/svntest/main.py
>> ===================================================================
>> --- subversion/tests/cmdline/svntest/main.py (revision 1885910)
>> +++ subversion/tests/cmdline/svntest/main.py (working copy)
>> @@ -129,7 +129,8 @@
>> if windows:
>> svneditor_script = os.path.join(sys.path[0], 'svneditor.bat')
>> else:
>> - svneditor_script = os.path.join(sys.path[0], 'svneditor.py')
>> + svneditor_script = sys.executable + " " + \
>> + os.path.join(sys.path[0], 'svneditor.py')
>>
>> # Username and password used by the working copies
>> wc_author = 'jrandom'
>>
>> ...which, somewhat to my surprise, seems to work :-), at least with
>> preliminary testing. I'd run it through the full test suite before
>> committing, of course.
>>
>> Another solution would be to create a 'svneditor.sh' that just runs
>> '${SVN_TEST_PYTHON} /path/to/svneditor.py $@'.
>
>Compare create_python_hook_script() — which, incidentally, isn't careful about
>using proper shell quoting where needed (or ensuring the value is safe to use
>unquoted).
>
>> I'm posting to get others' thoughts on how best to solve this. If the above
>> patch looks good, let me know and I'll test & commit it.
>
>+1 to using sys.executable. Fixing the quoting while in there would be nice to
>have, but as it's not a regression it's not a blocker either. Happy to leave
>the details to you.
+1. I'll look at create_python_hook_script(), make the necessary adjustments
to my patch (and maybe in create_python_hook_script() as well), and commit.
Thanks, Daniel.
Best regards,
-Karl
