[TIP] Tox command line substitutions

Sat Jan 8 13:11:26 PST 2011

Heh. Between then and this morning, I had an insight that made it easy to do.

So, the actual patch is ready.

However, because of two issues with the test suite as packaged, I'm
not 100% confident I've got a working patch. Maybe someone can offer
suggestions for how to get this to test? FWIW, it tests fine in Python
3.2b2

On Sat, Jan 8, 2011 at 1:56 PM, Kumar McMillan <kumar.mcmillan at gmail.com> wrote:
> On Fri, Jan 7, 2011 at 7:27 PM, Chris Rose <offline at offby1.net> wrote:
>> Actually in a CI system is exactly when I've used the positional
>> arguments; to add nose plugin configuration that enabled coverage and
>> xunit output :)
>>
>> This is turning out to be really, really tricky; the mini language for
>> substitution is not regular at all -- having to handle nested {}
>> constructs is giving me the twitch.
>>
>> basically, I'm stumped on how, exactly, to do this, without writing my
>> own lexer. I'm not even sure how to express the grammar.
>
> Oh, I see.  As a workaround you could make a bash script called
> run_tests.sh that did all the find ... -exec {} stuff and the
> nosetests regex stuff.  The bash script could use $@ to capture
> additional arguments and thus should work fine in tox as:
>
> [testenv]
> ...
> commands = run_tests.sh []
>
>>
>> On Fri, Jan 7, 2011 at 11:30 AM, Kumar McMillan
>> <kumar.mcmillan at gmail.com> wrote:
>>> On Fri, Jan 7, 2011 at 8:41 AM, Chris Rose <offline at offby1.net> wrote:
>>>> I'd like to propose one further amendment: Change from {} as a
>>>> substitution marker to {{}} (a al django templates, for example)
>>>
>>> maybe it would help to introduce another optional ini parameter?
>>>
>>> [testenv]
>>> raw_commands =
>>>    find ... -exec ... {}
>>>    run_tests.sh
>>>
>>> where raw_commands cannot be extended with additional command line
>>> parameters.  I think most people are using tox for CI so this might be
>>> a useful alternative when needed.
>>>
>>>> because I can think of at least one command pattern that's pretty
>>>> common (find ... -exec ... {} \;) that will trigger this.
>>>>
>>>> I'd call that a secondary concern, though. Getting the positional
>>>> argument syntax to be correct would be more important.
>>>>
>>>> On Fri, Jan 7, 2011 at 7:32 AM, holger krekel <holger at merlinux.eu> wrote:
>>>>> Hi Chris,
>>>>>
>>>>> just back from vacation - you also filed issue15 IISIC.
>>>>> (And i'd like to cross-post this mail to the issue but ASFAIK unfortunately
>>>>> GCode does not allow updating an issue by mail ...)
>>>>>
>>>>> http://code.google.com/p/pytox/issues/detail?id=15
>>>>>
>>>>> More inline commenting ...
>>>>>
>>>>> On Wed, Jan 05, 2011 at 08:48 -0700, Chris Rose wrote:
>>>>>> I ran into an issue with tox command positional argument substitution
>>>>>> that I was going to look into fixing, but it turns out that at least
>>>>>> to some extent the issue is by design (at least, according to the unit
>>>>>> tests in tox).
>>>>>>
>>>>>> Here's my own example, with expected behaviour:
>>>>>>
>>>>>> command = nosetests -m '(?:^|[\\b_\\./-])(?:unit)?[Tt]est'
>>>>>>
>>>>>> Tox substitutes this as:
>>>>>>
>>>>>> nosetests -m '(?:^|POSARGS)(?:unit)?POSARGSest'
>>>>>>
>>>>>> Obviously, this is undesirable.
>>>>>>
>>>>>> >From the unit tests, I see that by design tox is expected to support
>>>>>> replacing multi-word elements in the command as denoted by []
>>>>>> characters with the positional arguments, if provided, and if not, to
>>>>>> strip them out completely:
>>>>>>
>>>>>>            [section]
>>>>>>            key2=
>>>>>>                cmd1 []
>>>>>>                cmd2 [{item2} \
>>>>>>                     other]
>>>>>>
>>>>>> Here, cmd2 will either be "cmd2 POSARGS", "cmd2", or "cmd2 item2value"
>>>>>> depending on substitutions
>>>>>>
>>>>>> So, the question is, then, is this a good design for command
>>>>>> substitution? \[.*\] is a VERY greedy way to replace text in a
>>>>>> command, especially given that there's no way to escape these strings
>>>>>> as it stands. Also, any escaping scheme is going to be subject to some
>>>>>> potentially awful nested escaping issues, depending on how the command
>>>>>> is executed. That might not be a problem -- I suspect that the command
>>>>>> is executed using subprocess, so shell escaping is moot -- but it
>>>>>> might be a hard problem to solve.
>>>>>>
>>>>>> For my part, I'd suggest changing the substitution marker to something
>>>>>> less likely to ever be found in a command line; '{{}}' or '<<<>>>' or
>>>>>> something like it, or to add a way to escape them.
>>>>>
>>>>> To keep things mostly compatible i like your suggestion to
>>>>> (continue to) recognize "[]" literally and disallow specifying
>>>>> default arguments.  We should this via a a {*} substitution
>>>>> and phase out "[]" usage in the documentation and examples.
>>>>> This way we end upw ith only one substitution mechanism
>>>>> which makes it easier to think about escaping/providing
>>>>> default values etc.
>>>>>
>>>>>> I'd like to contribute a fix to this, but first I need to know what
>>>>>> the right fix should be. Thoughts?
>>>>>
>>>>> if the above makes sense i'd be happy to review and merge your fix.
>>>>>
>>>>> cheers,
>>>>> holger
>>>>>
>>>>>
>>>>>> --
>>>>>> Chris R.
>>>>>> ======
>>>>>> Not to be taken literally, internally, or seriously.
>>>>>> Twitter: http://twitter.com/offby1
>>>>>>
>>>>>> _______________________________________________
>>>>>> testing-in-python mailing list
>>>>>> testing-in-python at lists.idyll.org
>>>>>> http://lists.idyll.org/listinfo/testing-in-python
>>>>>>
>>>>>
>>>>> --
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Chris R.
>>>> ======
>>>> Not to be taken literally, internally, or seriously.
>>>> Twitter: http://twitter.com/offby1
>>>>
>>>> _______________________________________________
>>>> testing-in-python mailing list
>>>> testing-in-python at lists.idyll.org
>>>> http://lists.idyll.org/listinfo/testing-in-python
>>>>
>>>
>>
>>
>>
>> --
>> Chris R.
>> ======
>> Not to be taken literally, internally, or seriously.
>> Twitter: http://twitter.com/offby1
>>
>

-- 
Chris R.
======
Not to be taken literally, internally, or seriously.
Twitter: http://twitter.com/offby1