Why use send when you can call the method directly?

rubyonrails-talk
1

I am reading through some code in active_record/relation/query_methods - def build_where (about line 230); and there are a couple of calls to send. Here’s the first one

[@klass.send(:sanitize_sql, other.empty? ? opts : ([opts] + other))]

and here’s the second one

attributes = @klass.send(:expand_hash_conditions_for_aggregates, opts)

Why do we use send here when we could just call the method directly?

I mean, why is the first call not just

[@klass.sanitize_sql( other.empty? ? opts : ([opts] + other))]

Puzzled.

William

2

William Fisk wrote in post #977984:

I am reading through some code in active_record/relation/query_methods - def build_where (about line 230); and there are a couple of calls to send. Here's the first one [@klass.send(:sanitize_sql, other.empty? ? opts : ([opts] + other))]

and here's the second one attributes = @klass.send(:expand_hash_conditions_for_aggregates, opts)

Why do we use send here when we could just call the method directly? I mean, why is the first call not just [@klass.sanitize_sql( other.empty? ? opts : ([opts] + other))]

Puzzled.

The only reason I could see to do this would be to get around an access prohibition. You can use send in this way to call a private or protected method.

William

Best,

3

Ah, thanks, yes that will probably be the reason.

4

Ah, thanks, yes that will probably be the reason.

Right… send comes in real handy in testing private methods, use it all the time

5

David Kahn wrote in post #978011:

Ah, thanks, yes that will probably be the reason.

Right... send comes in real handy in testing private methods, use it all the time

Then you've got bigger problems. You shouldn't be testing your private methods; that's poking too deeply into an object's implementation. You should only ever test things that can be called from outside an object.

If a private method is an intermediate value in a computation, just test the end result. If a private method really needs to be tested separately, then it's telling you that it wants to be public.

Best,

6

The @klass variable also could == some_object.class If this is the case, then it could be a class method call, but still don't see the necessity to use #send.

7

David Kahn wrote in post #978011:

Ah, thanks, yes that will probably be the reason.

Right… send comes in real handy in testing private methods, use it all

the

time

Then you’ve got bigger problems. You shouldn’t be testing your private

methods; that’s poking too deeply into an object’s implementation. You

should only ever test things that can be called from outside an object.

If a private method is an intermediate value in a computation, just test

the end result. If a private method really needs to be tested

separately, then it’s telling you that it wants to be public.

I disagree – for me public is what I want to expose to outside access, regardless of complexity or size, where the private logic may actually be the most complex, being called by a relatively light public function. Now this may be a difference between TDD and BDD in pure forms, but for me I find if I drive my methods, public or private with test first, that the end resulting code is much more pliable. Also, if I have coverage on what becomes complex private logic, at a more granular level – then I have a much stronger project an also address minute issues closer to the source. I am sure you have your method of working that works for you, this is what I have found effective and successful.

8

David Kahn wrote in post #978200:

Then you've got bigger problems. You shouldn't be testing your private methods; that's poking too deeply into an object's implementation. You should only ever test things that can be called from outside an object.

If a private method is an intermediate value in a computation, just test the end result. If a private method really needs to be tested separately, then it's telling you that it wants to be public.

I disagree -- for me public is what I want to expose to outside access,

And therefore it is the only thing that is worth testing. No one cares what your private logic is like, because they never see it.

regardless of complexity or size, where the private logic may actually be the most complex, being called by a relatively light public function.

Then test the return from the "light" public function, or make your private logic public.

Now this may be a difference between TDD and BDD in pure forms, but for me I find if I drive my methods, public or private with test first, that the end resulting code is much more pliable.

No private method should have a test at all. That's testing implementation, and therefore is bad. You want to test interface only.

Also, if I have coverage on what becomes complex private logic, at a more granular level -- then I have a much stronger project an also address minute issues closer to the source.

No! All you have is brittle, implementation-dependent tests. What you *think* is improving the quality of your code is actually hindering it.

I am sure you have your method of working that works for you, this is what I have found effective and successful.

But it is neither. This is a matter of principles, not taste. Please do not handwave away what I am telling you by saying that it is a matter of taste -- these are principles you need to understand. If you want to disagree after you understand the principles, I will be interested to hear your reasoning.

It is *ineffective* to test private methods because you never care about the return value from a private method as such -- if you cared about the return value, you'd have made the method public.

It is *unsuccessful* to test private methods because you are tying your tests too closely to implementation, which means they will fail if you change your implementation -- this despite the fact that one of the points of testing is to tell you that your code still works when you refactor your implementation.

Best,

9

I would say that Marnen’s position is very consistent from what I see from Kent Beck and others on the XP/TDD lists. It took me a while to get used to the approach, and I think it’s important to clarify that Marnen isn’t saying not to test complex private methods. He’s just saying that if they are complex, maybe you’ve discovered another composed object’s public method. SRP and all that . . .

Best Wishes,

Peter

10

Peter Bell wrote in post #978204: [...]

I would say that Marnen's position is very consistent from what I see from Kent Beck and others on the XP/TDD lists.

The XP crowd have been a major influence on my testing philosophy, even though I've never done XP as such.

It took me a while to get used to the approach, and I think it's important to clarify that Marnen isn't saying not to test complex private methods.

Well, I guess I sort of *am* saying that, but only in the sense that you probably shouldn't *have* complex private methods that need their own tests -- that is, if your private methods are that complex, they should be made public and/or decomposed.

He's just saying that if they are complex, maybe you've discovered another composed object's public method.

Or the same object's public method, or a method that is too long...

SRP and all that . . .

I suppose. Basically, I think if you have a private method that is complex enough that wants to be tested in isolation, then that is a symptom of an underlying design problem: your methods are insufficiently accessible, or insufficiently atomic, or your code is otherwise insufficiently modular (say, by virtue of needing a new class introduced). Fix the underlying problem and the need for "so-big-it-needs-its-own-tests" private methods will go away.

Best Wishes, Peter

Best,