#respond_to? not working for dynamically generated methods

[maurice-gladwell]

On Wed, May 30, 2007 at 11:40:57PM +0900, Maurice G. wrote:

These are two completely different operations. The only relevant
solution here would be if you aspired to emulate Object#respond_to? by
checking if the call raises NoMethodError, but that’s infeasible because
method call commonly have side effects and/or are destructive.

I’m saying that if you’re doing

object.foo if object.respond_to? :foo

then you may as well do

object.foo rescue nil

(or a more specific rescue for NoMethodError)

I guess there are other uses for respond_to?, but I very rarely use it -
especially with ActiveRecord, which was the example given. Why do you
want
to check that an ActiveRecord class responds usefully to
find_by_foo_and_bar, but not actually call find_by_foo_and_bar?

Regards,

Brian.

[maurice-gladwell]

James G. wrote:

I don’t really understand this stance. My opinion is that providing
a method_missing() implementation is a convenient way for a
programmer to define a lot of dynamic methods. This increases the
messages an object responds to.

Following from that logic, I believe you should also override
respond_to?() to reflect those new messages. It’s my opinion
therefore that this thread has exposed a bug in Rails that should be
patched.

FWIW, there’s an entry about this in the Ruby Style Guide on
rubygarden.org:
http://wiki.rubygarden.org/Ruby/page/show/RubyStyleGuide/RespondToGoesWithMethodMissing

[maurice-gladwell]

Hi –

On Wed, 30 May 2007, Maurice G. wrote:

messages an object responds to.

I must agree with James Edward G. here.

David’s contrived example is of course possible, but highly unlikely in
practice; a programmer will generally not code his program into a
situation in which he can’t be sure whether an object will respond to a
particular message sent to it. That’s probably a major reason for having
Object#respond_to?.

But it can happen; hence NoMethodError.

In practice, and certainly in the Rails’ find_by_* case, you have a
well-defined range of messages to which an object will respond, known
already at development time. I don’t know of a single real-life example
in which a programmer doesn’t know at development time the entire range
of messages his objects might respond to. And Rails - with all its magic

• notwithstanding.

If it’s absolutely known what messages every object will respond to,
then we don’t need #respond_to? I don’t know… it’s obviously
just a slightly different way of looking at it. I’m uneasy with
having to rewrite respond_to? every time I implement method_missing;
it seems like a lot of work and an awfully tight coupling.

David

[maurice-gladwell]

Hi –

On Wed, 30 May 2007, Tim H. wrote:

patched.

FWIW, there’s an entry about this in the Ruby Style Guide on
rubygarden.org:
http://wiki.rubygarden.org/Ruby/page/show/RubyStyleGuide/RespondToGoesWithMethodMissing

Hmmm… perhaps there needs to be a second entry explaining why it’s
better not to put oneself in the position of having to maintain two
almost identical methods in parallel instead of one

David

[maurice-gladwell]

Hi –

On Wed, 30 May 2007, Maurice G. wrote:

David wrote:

If it’s absolutely known what messages every object will respond to,
then we don’t need #respond_to?

#respond_to? is usually used when we don’t know exactly which object is
referenced by a particular name, and especially what it can do (“duck
typing”). So we check if foo.respond_to? :to_s to see if we can tell
it to convert itself to a String.

#respond_to? != duck typing, though. At least “hard” duck typing is
just:

foo.to_s

But if we do know that foo is, say, an Array, we should - as the system
programmers - know which messages it will respond_to and how. It’s
generally never the case that we know which object we have, yet are not
sure which messages it will respond to.

Well… there you get into the issue of class vs. type. If you have a
spare six months, you can read the archives of this list on that topic
But certainly in ActiveRecord you get objects (association
collections) that will tell you that they are Arrays but which have
many methods beyond the vanilla Array.new object.

While you’re right that my rand(2) example is contrived, I’m not ready
to say that it’s so clear that method_missing can never be used to
handle cases that we couldn’t anticipate and enumerate. That seems
like a kind of retro-constraint, just so that it will interoperate
with respond_to?.

I don’t know… it’s obviously
just a slightly different way of looking at it. I’m uneasy with
having to rewrite respond_to? every time I implement method_missing;
it seems like a lot of work and an awfully tight coupling.

Yes, that is true. Everyone would be happy (some much happier, others a
tiny bit happier if we had a really working #respond_to? that would
cover all cases. But implementing such a thing elegantly seems
non-trivial.

I consider #respond_to? to be working, but I know what you mean

David

[maurice-gladwell]

On May 30, 2007, at 9:57 AM, Maurice G. wrote:

It would be interesting to see an implementation of James Edward
Gray’s
suggestion above. (“You use an Extract Method refactoring to pull out
the matching logic into a separate private method and use that in
both implementations.”)

#!/usr/bin/env ruby -wKU

class Test
def initialize(*fields)
@fields = fields.map { |f| f.to_s }
end

def find(*args)
“Performing query with args:\n#{args.inspect}”
end

def method_missing(meth, *args, &block)
if find_args = dynamic_finder?(meth)
find( find_args.first,
:conditions => find_args[1…-1].zip(args).
map { |n, v| “#{n} = ‘#
{v}’”}.
join(" AND ") )
else
super
end
end

def respond_to?(*args)
super || dynamic_finder?(args.first)
end

private

def dynamic_finder?(method_call)
if method_call.to_s =~ /\Afind(_all)?by(\w+)\Z/
args = [$1 ? :all : :first]
args += $2.split(“and”)
args if args[1…-1].all? { |f| @fields.include? f }
end
end
end

if FILE == $PROGRAM_NAME
model = Test.new(:name, :age, :height, :weight)

p !!model.respond_to?(:find_by_name_and_height)
p !!model.respond_to?(:find_all_by_age)
p !!model.respond_to?(:find_all_by_race)

puts

puts model.find_by_name_and_height(“James”, “61 in.”)
puts model.find_all_by_name_and_height
end

James Edward G. II

[maurice-gladwell]

David wrote:

If it’s absolutely known what messages every object will respond to,
then we don’t need #respond_to?

#respond_to? is usually used when we don’t know exactly which object is
referenced by a particular name, and especially what it can do (“duck
typing”). So we check if foo.respond_to? :to_s to see if we can tell
it to convert itself to a String.

But if we do know that foo is, say, an Array, we should - as the system
programmers - know which messages it will respond_to and how. It’s
generally never the case that we know which object we have, yet are not
sure which messages it will respond to.

I don’t know… it’s obviously
just a slightly different way of looking at it. I’m uneasy with
having to rewrite respond_to? every time I implement method_missing;
it seems like a lot of work and an awfully tight coupling.

Yes, that is true. Everyone would be happy (some much happier, others a
tiny bit happier if we had a really working #respond_to? that would
cover all cases. But implementing such a thing elegantly seems
non-trivial.

It would be interesting to see an implementation of James Edward G.'s
suggestion above. (“You use an Extract Method refactoring to pull out
the matching logic into a separate private method and use that in
both implementations.”)

–
M.

[maurice-gladwell]

On 5/30/07, James Edward G. II [email protected] wrote:

class Test
find( find_args.first,
super || dynamic_finder?(args.first)
end

puts model.find_by_name_and_height(“James”, “61 in.”)
puts model.find_all_by_name_and_height
end

James Edward G. II

This is very interesting, I just imagine that dynamic_finder does not
give consistent results depending on events outside the Ruby
program(*). Are you still happy with the semantics, instead of
respond_to? giving false negatives it gives false positives now.
A second issue was brought up by David,( I have no idea why he put a
smiley
You have created a dependency between method_missing and respond_to? I
really feel bad about this.

Robert

(*) or maybe by threading

[maurice-gladwell]

James Edward G. II schrieb:

What exactly is the use case for respond_to?()?

Only as a remark for a special usage, which is not easy (impossible?)
without
“respond_to?()”.

I have some Libraries for textual analysis, which are usually used via
“irb”.
The application works on large Arrays of String objects. Based on
regular
expressions I define sometimes methods for some of these object, during
one
textual analysis often several methods.

Then I can categorize the String objects in the Array by using
“respond_to?()”.

It is very easy and helpful (may be it sounds complex an strange, but
this is
only my bad English).

Wolfgang Nádasi-Donner

[maurice-gladwell]

On 5/30/07, James Edward G. II [email protected] wrote:

I don’t really understand this stance. My opinion is that providing
respond_to?():
methods, so the fact that respond_to?() returns false for them makes
the above documentation lie.
James the documentation is not correct anyway, look at the wording,
nothing ever can reply to methods, only to messages. I liked a lot
what David has said about this. >That does not seem good to me and
it’s
certainly possible to fix it. Thus, I can only reason that it is a
bug in ActiveRecord.
That is kind of a thing I have not considered in my reply, I
misunderstood that you were dreaming about a magical
Object#respond_to? which would interpret the semantics of
#method_missing generally, hence my surprise.

Is this really a good idea?

Is it a good idea not to? What exactly is the use case for
respond_to?()? I have only ever used it to find out if something I
want is available. If I can’t do that reliably, it doesn’t really
help me much.
Hmm sure but I feel its need more in MetaProgramming, if you want to
fix ActiveRecord I cannot argue with you at all I do not know that
stuff, so I just bail out of the discussion.

However if you would like to spend some time in philosophy about
respond_to? I see a clearcut dilemma between the ideal behavior of
respond_to? and method_missing. See my example above but it is easy to
imagine that #method_missing is dispatching depending on an external
datasource.
How would you want #respond_to? to handle this?

I just feel that for practical reasons #respond_to? does a good job
already. This might
Robert

[maurice-gladwell]

On May 30, 2007, at 3:31 PM, Robert D. wrote:

This is very interesting, I just imagine that dynamic_finder does not
give consistent results depending on events outside the Ruby
program(*). Are you still happy with the semantics, instead of
respond_to? giving false negatives it gives false positives now.

Well, if it can’t be made accurate, that’s a different thing. First,
we would probably need to document that it can’t be trusted in such a
case.

You have created a dependency between method_missing and respond_to? I
really feel bad about this.

I guess that means that I made it so that respond_to?() needs to be
updated if method_missing() is? Yes, I guess that’s right.

David says that’s bad. OK. I thought respond_to?() lying to me was
bad. So I guess we just chose to care about different things.

I guess each person needs to decide what is more important to them.

James Edward G. II

[maurice-gladwell]

On 5/30/07, James Edward G. II [email protected] wrote:

You have created a dependency between method_missing and respond_to? I
really feel bad about this.

I guess that means that I made it so that respond_to?() needs to be
updated if method_missing() is? Yes, I guess that’s right.

David says that’s bad. OK.
And so do I, well that sounds quite prepotent, I know.
I thought respond_to?() lying to me was
bad. So I guess we just chose to care about different things.

I guess each person needs to decide what is more important to them.
Well I was hoping to convince you, sure I was, but I see now that
after having discussed things we just have different opinions and that
is that of course.

The most important point I wanted to make - and I probably failed - is
that I am not sure if it really lies for me, but this is a tricky
issue and I can understand what you dislike about it.

Cheers
Robert

[maurice-gladwell]

On 5/30/07, Wolfgang Nádasi-Donner [email protected] wrote:

Then I can categorize the String objects in the Array by using “respond_to?()”.

It is very easy and helpful (may be it sounds complex an strange, but this is
only my bad English).
Hmm as fellow native German Speaker I think your English is perfect.
Could you give an example?
I somehow feel that you might use a different approach than
hitchhiking methods, or are the used for different purposes too.

Sounds interesting.

Cheers
Robert

[maurice-gladwell]

On 5/31/07, Wolfgang Nádasi-Donner [email protected] wrote:

Robert D. schrieb:

Sounds interesting.

I don’t know, because it is a special situatuation, and it is very experimental.

Well it still sounds very interesting to me, I remember when I was
writing Emacs Macros for jumping between events in very very long log
files. If I guess correctly what you are doing here is very intersting
and quite complicated.
But to come back to respond_to?
Am I guessing correctly that you use respond_to? in order to manage
the abilities you have already created for your singletons, as these
abilities are methods?

Cheers
Robert

[maurice-gladwell]

Robert D. schrieb:

Am I guessing correctly that you use respond_to? in order to manage
the abilities you have already created for your singletons, as these
abilities are methods?

Yes, these methods are something like “executable attributes” - but - I
would
prefer in the moment to say: “don’t use it in a real program!”.

The usage background is as already named:

1. Execution speed does not matter (even one minute is not a problem),
2. readability does not matter (there are no comments at all),
3. size of code to write does matter (less is better).

I will not recommend to write software (=real programs with a life
cycle) based
on this principles.

Wolfgang Nádasi-Donner

[maurice-gladwell]

Hi –

On Thu, 31 May 2007, James Edward G. II wrote:

You have created a dependency between method_missing and respond_to? I
really feel bad about this.

I guess that means that I made it so that respond_to?() needs to be updated
if method_missing() is? Yes, I guess that’s right.

David says that’s bad. OK. I thought respond_to?() lying to me was bad. So
I guess we just chose to care about different things.

I just raise an eyebrow at that much repeated/parallel code.

I guess each person needs to decide what is more important to them.

I think it’s just a matter of how you define respond_to? conceptually
– as a handle on the methods in the object’s method lookup path, or
as a registry of everything the object can do without raising an
error. It’s interesting in this connection that delegation adds the
delegated methods to the respond_to? roster. I guess that summarizes
the difference for me: a method that’s delegated is known to be taken
care of, whereas a method that’s missing is missing (even if the
potential error can be averted).

David

[maurice-gladwell]

Robert D. schrieb:

Sounds interesting.

I don’t know, because it is a special situatuation, and it is very
experimental.
During usage i build something like a hybride human-software process. At
start
some helper libraries are required in “irb”, and then I worked with
several data
in different buffers, writing helpful code for analysis and changes and
throw
the code away after the work is done (usually there is no future use for
the
code). If one wants to use this in a “real program” s/he must find out
how to
generate (=create) useful Ruby code based on some events - I mean code,
that is
not predictable when the program starts.

Robert D. schrieb:

Could you give an example?

This is difficult, because there is no fixed code, but I can decribe a
typical
process.

Someone comes to me with a CD (or ZIP file via mail) with several
textual files,
and says “…something is wrong there…”, or “…can you help me by
producing a
short overview…”. Especially in the first situation I have nearly no
addition
information. So I load one File into a line oriented buffer structure
(based on
an Array of Strings) an start analysing the data by using helper methods
and
regular expressions.

Sometimes (=often) the files have explicit relationships via textual
remarks.
This is one case, where the usage of singleton methods for lines
(=String
objects in an Array) comes up, because it is simple and easy to use. It
follows
usually the same pattern:

buffer.each do |line|
if (md = line.match(/a pattern that is of some interest/)
# define a special method for the line
end
end

The definition will take place on the assumption, that the method may be
useful
later on, which is not clear in the moment. It is possible, that the
method
opens a new buffer, loads another file and does some action on it.

This can happen several times. Later it is possible that I want, that a
buffer
applyies such a method for each line the method is defined. I can use
the
following short pattern for this:

buffer.each{|line|line.my_method if line.respond_to?(:my_method)}

Remarks:

1. Performance isn’t a interesting thing in this situation. When I think
   about
   an output and plan what to do next, I am the bottle-neck in the system.

2. This kind of using singleton methods is attractive, because it isn’t
   complicated and simple to use (=not much to write), what is very
   interesting in
   an interactive session.

3. Readability of the code isn’t relevant, because it will be thrown
   away after
   usage. This can be done, because the problems differ very much. Reusable
   things
   will be build in the libraries, that I require in the very beginning.

4. Readability isn’t a big problem, because I wrote some large TECO
   programs
   long time ago…

I Think this is a very special usage of Ruby.

Wolfgang Nádasi-Donner