Feature #15554: warn/error passing a block to a method which never use a block

• Status: Open
• Priority: Normal
• Assignee: matz (Yukihiro Matsumoto)
• Target version: -

[ko1 (Koichi Sasada)]

Abstract

Warn or raise an ArgumentError if block is passed to a method which does not use a block.
In other words, detect "block user methods" implicitly and only "block user methods" can accept a block.

Background

Sometimes, we pass a block to a method which ignores the passed block accidentally.

def my_open(name)
  open(name)
end

# user hopes it works as Kernel#open which invokes a block with opened file.
my_open(name){|f| important_work_with f }
# but simply ignored...

To solve this issue, this feature request propose showing warnings or raising an exception on such case.

Last developer's meeting, matz proposed &nil which declares this method never receive a block. It is explicit, but it is tough to add this &nil parameter declaration to all of methods (do you want to add it to def []=(i, e, &nil)?).
(I agree &nil is valuable on some situations)

Spec

Define "use a block" methods

We need to define which method accepts a block and which method does not.

• (1) method has a block parameter (&b)
• (2) method body has `yield'
• (3) method body has super (ZSUPER in internal terminology) or super(...)
• (4) method body has singleton method (optional)

(1) and (2) is very clear. I need to explain about (3) and (4).

(3). super (ZSUPER) passes all parameters as arguments. So there is no surprise that which can accept block.
However super(...) also passes a block if no explicit block passing (like super(){} or super(&b)) are written.
I'm not sure we need to continue this strange specification, but to keep compatibility depending this spec, I add this rule.

(4). surprisingly, the following code invoke a block:

def foo
  class << Object.new
    yield
  end
end

foo{ p :ok } #=> :ok

I'm also not sure we need to keep this spec, but to allow this spec, I added (4) rule.
Strictly speaking, it is not required, but we don't keep the link from singleton class ISeq to lexical parent iseq now, so I added it.

Exceptional cases

A method called by super doesnt warn warning even if this method doesn't use a block.
The rule (3) can pass blocks easily and there are many methods dont use a block.

So my patch ignores callings by super.

corner cases

There are several cases to use block without (1)-(4) rules.

Proc.new/proc/lambda without a block

Now it was deprecated in r66772 (9f1fb0a17febc59356d58cef5e98db61a3c03550).
Related discussion: [Bug #15539]

block_given?

block_given? expects block, but I believe we use it with yield or a block parameter.
If you know the usecase without them, please tell us.

yield in eval

We can't know yield (or (3), (4) rule) in an eval evaluating string at calling time.

def foo
  eval('yield`)
end

foo{} # at calling time,
      # we can't know the method foo can accept a block or not.

So I added a warning to use yield in eval like that: test.rb:4: warning: use yield in eval will not be supported in Ruby 3.

Workaround is use a block parameter explicitly.

def foo &b
  eval('b.call')
end

foo{ p :ok }

Implementation

Strategy is:

• [compile time] introduce iseq::has_yield field and check it if the iseq (or child iseq) contains yield (or something)
• [calling time] if block is given, check iseq::has_yield flag and show warning (or raise an exception)

https://gist.github.com/ko1/c9148ad0224bf5befa3cc76ed2220c0b

On this patch, now it raises an error to make it easy to detect.
It is easy to switch to show the warning.

Evaluation and discussion

I tried to avoid ruby's tests.

https://gist.github.com/ko1/37483e7940cdc4390bf8eb0001883786

Here is a patch.

There are several patterns to avoid warnings.

tests for block_given?, Proc.new (and similar) without block

Add a dummy block parameter.
It is test-specific issue.

empty each

Some tests add each methods do not yield, like: def each; end.
Maybe test-specific issue, and adding a dummy block parameter.

Subtyping / duck typing

https://github.com/ruby/ruby/blob/c01a5ee85e2d6a7128cccafb143bfa694284ca87/lib/optparse.rb#L698

This parse method doesn't use yield, but other sub-type's parse methods use.

super with new method

https://gist.github.com/ko1/37483e7940cdc4390bf8eb0001883786#file-tests-patch-L61

This method override Class#new method and introduce a hook with block (yield a block in this hook code).

https://github.com/ruby/ruby/blob/trunk/lib/rubygems/package/tar_writer.rb#L81

In this method, call super and it also passing a block. However, called initialize doesn't use a block.

Change robustness

This change reduce robustness for API change.

Delegator requires to support __getobj__ for client classes.
Now __getobj__ should accept block but most of __getobj__ clients do not call given block.

https://github.com/ruby/ruby/blob/trunk/lib/delegate.rb#L80

This is because of delegator.rb's API change.

https://gist.github.com/ko1/37483e7940cdc4390bf8eb0001883786#file-tests-patch-L86

Nobu says calling block is not required (ignoring a block is no problem) so it is not a bug for delegator client classes.

Found issues.

[ 2945/20449] Rinda::TestRingServer#test_do_reply = 0.00 s
  1) Error:
Rinda::TestRingServer#test_do_reply:
ArgumentError: passing block to the method "with_timeout" (defined at /home/ko1/src/ruby/trunk/test/rinda/test_rinda.rb:787) is never used.
    /home/ko1/src/ruby/trunk/test/rinda/test_rinda.rb:635:in `test_do_reply'

[ 2946/20449] Rinda::TestRingServer#test_do_reply_local = 0.00 s
  2) Error:
Rinda::TestRingServer#test_do_reply_local:
ArgumentError: passing block to the method "with_timeout" (defined at /home/ko1/src/ruby/trunk/test/rinda/test_rinda.rb:787) is never used.
    /home/ko1/src/ruby/trunk/test/rinda/test_rinda.rb:657:in `test_do_reply_local'

[10024/20449] TestGemRequestSetGemDependencyAPI#test_platform_mswin = 0.01 s
  3) Error:
TestGemRequestSetGemDependencyAPI#test_platform_mswin:
ArgumentError: passing block to the method "util_set_arch" (defined at /home/ko1/src/ruby/trunk/lib/rubygems/test_case.rb:1053) is never used.
    /home/ko1/src/ruby/trunk/test/rubygems/test_gem_request_set_gem_dependency_api.rb:655:in `test_platform_mswin'

[10025/20449] TestGemRequestSetGemDependencyAPI#test_platforms = 0.01 s
  4) Error:
TestGemRequestSetGemDependencyAPI#test_platforms:
ArgumentError: passing block to the method "util_set_arch" (defined at /home/ko1/src/ruby/trunk/lib/rubygems/test_case.rb:1053) is never used.
    /home/ko1/src/ruby/trunk/test/rubygems/test_gem_request_set_gem_dependency_api.rb:711:in `test_platforms'

These 4 detection show the problem. with_timeout method (used in Rinda test) and util_set_arch method (used in Rubygems test) simply ignore the given block.
So these tests are simply ignored.

I reported them. (https://github.com/rubygems/rubygems/issues/2601)

raise an error or show a warning?

At least, Ruby 2.7 should show warning for this kind of violation with -w.
How about for Ruby3?

[ko1 (Koichi Sasada)]

• Description updated (diff)

[hsbt (Hiroshi SHIBATA)]

I fixed the issues of TestGemRequestSetGemDependencyAPI on upstream repository and merged them at r66904.

[alanwu (Alan Wu)]

Related: Feature #10499

[decuplet (Nikita Shilnikov)]

I have a nice example where I use calls like super { ... } even if the super method doesn't yield a block. From my understanding, this behavior won't be broken by the changes, but I still want to add it to the context.
I'm the author of the dry-monads gem, it defines, you might have guessed, a bunch of monads. For instance, there is the Result monad which represents possibly-unsuccessful computation. A typical use case:

def call(params)
  validate(params).bind { |values|
    create_account(values[:account]).bind { |account|
      create_owner(account, values[:owner]).fmap { |owner|
        [account, owner]
      }
    }
  }
end

Here validate, create_account, and create_owner all return a Result value, binds compose the results together. That's rather ugly, that's why the gem also adds so-called do notation (the name has stolen from Haskell), it's a mixin you add to a class which prepends every method you define and passes a block which in order tries to unwrap a Result value. Better see the code and read the comments:

# this line also prepends the current class with a dynamically created module
include Dry::Monads::Do

# this method will be overridden in the prepended module
def call(params)
  # yield will halt the execution if validate returns Failure(...)
  # or unwraps the result if it returns Success(...) so that it'll
  # be assigned to values. This way we avoid the chain of .bind calls
  values = yield validate(params)
  account = yield create_account(values[:account])
  owner = yield create_owner(account, values[:owner])

  Success([account, owner])
end

Obviously, the do version is way cleaner yet has the same semantics. Dry::Monads::Do wraps all the methods as they are defined so that you don't need to worry whether a method unwraps Results or not. If it doesn't, then it just doesn't call the block, that simple. Also Dry::Monads::Do is smart enough to discard its block if another block is given to a method:

include Dry::Monads::Result::Mixin
include Dry::Monads::Do

def foo
  bar { 5 }
  baz # baz will return 6, see below
end

def bar
  # here the block comes from foo, not from Do
  yield + 1
end

def baz
  # no block from foo given, Do in action
  yield Success(6)
end

Now here's the problem.

include Dry::Monads::Do

def call
  foo
end

def foo
  # foo gets the block from Do but doesn't use it
  5
end

In the last example, I have no means to detect that foo won't use the block. I could analyze parameters, I guess it would be slower, but if a method uses yield I just cannot detect it without analyzing the source code, that'd be dead slow and not reliable either.

See what Do does here and there for more details.

I should add that dry-monads is quite popular, it's not only used by me :) (see it on rubygems https://rubygems.org/gems/dry-monads)

---

Gem's docs

[localhostdotdev (localhost .dev)]

To detect if a block is used, binding would also need to be detected, e.g.: def b(arg); arg.eval("yield"); end; def a; b(binding); end

[matthewd (Matthew Draper)]

This is great! Ignored blocks can be very confusing.

A method called by super doesnt warn warning even if this method doesn't use a block.
The rule (3) can pass blocks easily and there are many methods dont use a block.

So my patch ignores callings by super.

Would it be possible for a method called by super to only ignore an unexpected block if the calling method (that contained super) also contained something that explicitly uses/consumes the block (&b / yield)?

I think that would still allow existing uses where a method consumes a block itself, and then (accidentally) implicitly passes it to a super that doesn't want it -- but that it would also keep the warning in the other situation, where an entire super-chain does not use the block.

Otherwise I am worried that super becomes a bad thing to use, because it completely disables this new safety feature.

[k0kubun (Takashi Kokubun)]

• Description updated (diff)

[k0kubun (Takashi Kokubun)]

• Copied from Feature #10499: Eliminate implicit magic in Proc.new and Kernel#proc added