[Freeipa-devel] [PATCHES] 0094-0099 Make Fuzzy objects available for the whole ipatests module

[Rob Crittenden]

Petr Viktorin wrote:
> On 09/13/2013 07:04 PM, Rob Crittenden wrote:
>> Petr Viktorin wrote:
>>> On 09/13/2013 04:12 PM, Tomas Babej wrote:
>>>> Hi,
>>>>
>>>> The following patches move and extend the functionality of Fuzzy
>>>> objects. This is necessary to use them from within integration tests.
>>>
>>> -1 to the idea.
>>> I'm not a fan of the Fuzzy objects; they basically exist so that you can
>>> use regex matching in the Declarative tests.
>>> As you've probably noticed Fuzzy is quite specific to the framework and
>>> its test suite -- see the strict bytes/unicode separation and the amount
>>> of changes it takes to tear them out.
>>>
>>> I'm also not a fan of having a generic "Match anything using some rules"
>>> object where everybody adds behavior specific to their use case. Each
>>> addition increases the complexity and number of corner cases, and the
>>> complete intended functionality can never be achieved.
>>>
>>> Using regular expressions directly should actually be easier and less
>>> error-prone in most cases.
>>> If you disagree, I'd like to see your use case.
>>
>> I'm not sure what the objection is, Fuzzy is just an abstraction. It
>> lets one do in-line regex which is the reason it was introduced
>> (initially for uid and gid because they are non-deterministic).
>
> Yes. "In-line" is the key here. It lets you do regex matching via the ==
> operator, which is what Declarative tests need, but elsewhere the
> abstraction is completely unnecessary.
>
>> Replacing it would either a) replicate its functionality almost
>> completely or b) spread duplicate regex code all over the place.
>
> I'd go for b; spreading this code:
>      import re
>      some_regex = re.compile('some.*regex$')
>      ...
>      assert some_regex.search(x)
> instead of:
>      from wherever import Fuzzy
>      warm_fuzzy = Fuzzy('some.*regex$')
>      ...
>      assert x == warm_fuzzy
> all over the place is fine in my book. And you can even, say, add custom
> flags to the regex without complicating shared code.

Right, and it's all duplicate. Just look in his patch how many times 
fuzzy.digits is used. What's going to happen is someone is going to come 
along later and say "Geez, we have a ton of some_regex = 
re.compile('\d+'), I should make a macro thinger out of this" and we're 
back where we started.

> The rest of Fuzzy functionality consists of strict type checking (which
> isn't really necessary in integration tests), and the ability to call
> arbitrary callables (which is just the scope creep I was talking about).
> By the way, in current tests these features are hardly ever used in
> combination.

Here we agree. I'd prefer to keep Fuzzy limited to just simple regex and 
woudln't object to delegating the other enforcement to something else.

> Even if this extra functionality is necessary, it's orthogonal to regex
> matching and can be more cleanly done with several separate asserts.
> Unless you need a single declarative object to compare against, of course.

Yup.

>
>> That isn't to say that Fuzzy isn't being abused, but that is also the
>> responsibility of the reviewers to be strict about.
>
> Then perhaps I'm too strict, but I say that using it outside of the
> declarative tests is abuse.
> Especially if it takes six patches with hundreds of changed lines to
> repurpose Fuzzy for integration tests (but that's not part of "-1 to the
> idea").

That is hardly fair. The bulk of the patches just change imports.

So to summarize, I think:

- Fuzzy should remain as a regex should remain as a regex shortcut
- The non-regex features can be moved elsewhere
- I don't really have a handle on how he intended to use this for 
integration testing, so I don't have an opinion here. But I'd expect 
that most integration tests depend more on return values than comparing 
against "known good".

rob