DevelopersMeeting20201210Japan

https://bugs.ruby-lang.org/issues/17346

DateTime and location

12/10 (Thu) 13:00-17:00 JST @ Online

Next Date

2021/01/13 (Wed) 13:00-
- pre-reading: 2021/01/12 (Tue) 13:00-

Announce

About 3.0 timeframe

preview 3 or rc1
- Need to try removal of WEBrick

3.0 urgent tickets

[Feature #17371] Reintroduce `expr in pat` (ktsj)

Is it OK?

Preliminary discussion:

The proposal:

p(1 in a) #=> true (side effect: a = 1)
p(1 in 2) #=> false

Discussion:

matz: not against it basically, but do you want it in 3.0?
k-tsj: I want to remove experimental flag for pattern match features as a whole in 3.1. In order to do so we need to add experimental features this month.

Conclusion:

matz: go ahead as an experimental feature

The proposal: R-assign should return the assigned value?

1 => a  #=> 1 (return the assigned value)

nobu: Re: value of =>, is this okay?

1=>a # r-assign
p(1=>a) # hash: p(**{1 => a})

Conclusion:

ktsj: postpone it. we can continue to discuss towards 3.1

ktsj: How about allowing the typical r-assign case as an official (not experimental) feature?

1 => [a, b] #=> warning: experimental
1 => a      #=> no warning (only if the target is one variable)

# master: warning: One-line pattern matching is experimental,
#                  and the behavior may change in future versions of Ruby!

Conclusion:

matz: go ahead

[Bug #17369] Introduce non-blocking `Process.wait`, `Kernel.system` and related methods.

ko1: Process::Status.wait has been merged. Okay?
nobu: concerns:
- is the name okay?
- is it okay to change existing CAPI?
- it can break marshal result of Process::Status

About naming

matz: The name seems OK
matz: It’s not OK to change C API
akr: Ideally we want a PID class.

About hooks

mame: adding a hook is okay?
nobu: It can’t be helped

About data structure

akr: error should not be a part of the struct. the status doesn’t exist when it fails. (Also, Ruby 2.7 has no error in Process::Status.)

About marshalling

nobu: This may break Marshal compatibility because it changes the representation to T_DATA?
akr: I confirm the incompatibility: T_OBJECT -> T_TYPEDDATA

Conclusion:

matz: In general, I accept the feature. The name Process::Status.wait is okay. I want nobu to review, shepherd, and re-implement (if needed)
naruse: I’ll revert if the implementation does not meet the deadline
nobu: will do soon
matz: naruse-san, let me know if the rc1 date is fixed

[Feature #17303] Make webrick to bundled gems or remove from stdlib (hsbt)

duerst: Does it remain as a normal gem?
naruse: Yes.
a_matsuda: Rails uses capybara, and I wonder capybara can use webrick…
naruse: Is it hard for capybara to use what rails uses (puma?)
a_matsuda: … or let it explictly depend on webrick. Both are possible.
a_matsuda: Found that rack uses webrick. It is autoloaded.

Conclusion:

matz: OK to remove as long as we have one more RC before the final 3.0.
mame: nobu, please deal with un.rb that currently depends on webrick
need to handle rubygems (-> hsbt) and rdoc (-> aycabta)

[Feature #17351] Deprecate Random::DEFAULT (eregon)

Is it OK?

Preliminary discussion:

ko1: can we put Class for it?
nobu: it should be something rather than Class
mame: Here was the original motivation of Random::DEFAULT

def foo(random: Random::DEFAULT)
end

ko1: it can be rewrite with.

def foo(random: Random)
end

mame: However, we cannot find this use case by gem-codesearch
mrkn: If we need to do so, now we can use Random itself
ko1: How can we deprecate it? Warn it when it is accessed?
nobu: Use deprecate_constant

Discussion:

akr: compatible with older versions?
nobu: From 1.9, it is available. Only Random.seed is introduced from 3.0.

Conclusion:

matz: go ahead.

[Feature #17323] Ractor-local storage (marcandre)

What API to use?

Discussion:

ko1: how about LVar#per_ractor_value instead of LVar#value ?

CACHE = Ractor::LVar.new{ {} }

# rejected because user expects unique 
p CACHE.value #=> {}

# new proposal
CACHE.per_ractor_value[:a] = 1

Ractor.new{
  CACHE.per_ractor_value #=> {} empty hash
}

# like Thread#[]

Ractor.current[:cache] = {:a}

r = Ractor.new do
  Ractor.current[:cache] = {:b}
end.take

p Ractor.current[:cache] #=> {:a}

# With special key instead of Symbol literals

CACHE_KEY = Symbol.new
Ractor.current[CACHE_KEY] = {:a}

r = Ractor.new do
  Ractor.current[CACHE_KEY] = {:b}
end.take

p Ractor.current[CACHE_KEY] #=> {:a}

# current avaiable way to handle Ractor local variables

p Thread.main
#=> #<Thread:0x000002992d6fca98 run>

p Ractor.new{ Thread.main }.take
#=> #<Thread:0x0000029932cb1c68 dead>

Thread.main[:key] = 1
Ractor.new{
  p Thread.main[:key] = 2 #=> 2
}.take
p Thread.main[:key]       #=> 1

Conclusion:

matz: accept Ractor#[sym] as Thread#[sym]. Symbol.new will be discussed for Ruby 3.1.

[Feature #17378] Ractor#receive with filtering like other actor langauge (ko1)

I could not find this simple pattern description.

Discussion:

(very long explanation with some slides by ko1)

Conclusion:

matz: go ahead as experimental (Ractor.receive_if{ ... })

report: rb_ext_ractor_safe(bool flag)

Now all extensions are ractor unsafe.


Init_foo(void)
{
  rb_ext_ractor_safe(true);

  rb_define_method(...); // ractor-safe method (can run on other Ractors)
}

normal tickets

[Feature #16697] Hash.ruby2_keywords_hash?(value) should support any object (eregon)

Yes/no?

Preliminary discussion:

Hash.ruby2_keywords_hash?("str")
#=> expected: false, 
#=> actual: TypeError

ruby2_keywords def foo(*args)
  if args.last.is_a?(Hash) && Hash.ruby2_keywords_hash?(args.last)
  else
  end
end

mame: not negative
nobu: look good

Discussion:

shyouhei: is ruby2_keywords def foo(*args) the exepcted usage?
mame: yes.
akr: why is this needed?
mame: use case: https://github.com/ruby/ruby/pull/2966/files#r394741112
nobu: I don’t bother this behaviour.
mame: I cannot understand the use case, but I’m okay for the change. I don’t know if this is backported to 2.7 or not
mame: ah, if 2.7 and 3.0 have different semantics of Hash.ruby2_keywords_hash?, it may bring another trouble. Maybe we should ask nagachika-san if he backport this change to 2.7 or not
mame: umm. I’m now afraid if this change brings new confusion. It is trivial to workaround the issue.

Conclusion:

mame: I’ll ask nagachika-san

[Feature #17055] Allow suppressing uninitialized instance variable and method redefined verbose mode warnings (jeremyevans0)

Can we completely remove uninitialized instance variable warnings? Then we don’t need this feature. I would very much like that in Ruby 3.

Preliminary discussion:

mame: Agreed
ko1: It can speed up optcarrot a bit.

Discussion:

nobu: I’m unsure if the warning is helpful. It might be, but I can’t remember
ko1: thought it was dangerous but on a second thought I now think we can give it a try.
a_matsuda: I don’t think anybody complains about it immediately (people will not be aware of this if warnings disappear).
a_matsuda: I don’t remember any typo saved by this though.
a_matsuda: Maybe rubocop remains as-is so ruby side warnings might not be that important.

Conclusion:

matz: go ahead

[Bug #17280] Dir.glob with FNM_DOTMATCH matches “..” and “.” and results in duplicated entries (jeremyevans0)

Do we want to change the behavior to exclude “..” and duplicate results?
If so, is the pull request acceptable?

Preliminary discussion:

The following seems to have duplicated entries like “bar” and “bar/.”.

This is because the recursive match (**/) removes .. but not .

remove /. too?

% ruby -e 'p Dir.glob("**/*", File::FNM_DOTMATCH)'
[".", "bar", "bar/.", "bar/.baz", "bar/.baz/.", "bar/.baz/qux"]

The following “..” seems unexpected?
This is because ** is not a recursive match
- ** is equal to *, which is a bit confusing. A warning will solve the issue?
```
% ruby -e 'p Dir.glob("**", File::FNM_DOTMATCH)' 
[".", "..", "bar"]
```
This is expected?

% ruby -e 'p Dir.glob("*", File::FNM_DOTMATCH)' 
[".", "..", "bar"]

mame: I think we can postpone it to Ruby 3.1. Any hasty change looks unreasonably risky to me.

Discussion:

akr: Is “.” returned or not?
nobu: zsh seems to have similar behaviour.
matz: “don’t repeat a file twice” doesn’t sound a bad thing to me.
akr: what is “a file”? it’s unclear. “bar” and “bar/.” is the same (hard-linked) file. If we specify Dir.glob don’t repeat hard-linked files, hard-linked files other than “.” and “..” will also be unappeared. Another specification is “don’t return . and ..” (like zsh). However it changes Dir.glob doesn’t return “.”, “.” is not repeated, though.
nobu: /. is in fact a hard link.
znz: it seems zsh special-cases . and .. regardless of GLOB_DOTS.

Conclusion:

matz: try to remove “.” and “..” on 3.1 (early timing to check compatibility issues)

[Bug #17162] Dir[‘*/’] : stack smashing detected when listing big amount of directories (jeremyevans0)

Do we want to add an internal recursion limit to try to prevent this?

Preliminary discussion:

nobu: The issue is not Dir.[] but stack overflow detection bug.
mame: If so, the limitation is not needed

Discussion:

ko1: it is impossible to detect machine stack overflow.

Conclusion:

mame: let’s postpone it to 3.1
matz: agreed

[Bug #17218] Range#step sometimes behaves unexpectedly with Rational endpoints and increment (jeremyevans0)

This is caused by calling rb_int_* functions directly instead of Ruby methods.
Do we want to switch arith_seq_last to using Ruby methods, and if so, should all of the rb_int_* be switched, or just the minimal number?
The pull request only changes three of the rb_int_* to rb_funcall, is it OK?

Preliminary discussion:

mrkn: will look

Conclusion:

mkrn: already fixed

[Bug #13663] `String#upto` doesn’t work as expected (jeremyevans0)

I think the current behavior of String#upto in this case is expected and not a bug.
Is it OK to close this?

Preliminary discussion:

'x'.upto('ac').to_a #=> []
# expectation: ["x", "y", "z", "aa", "ab", "ac"]

mame: agreed with closing

Conclusion:

matz: ok to close.

[Feature #17342] Add Hash#fetch_set (maxlap)

Yes/no?

Preliminary discussion:

mame: I think the focus is whether we want to use false or nil for this code pattern
nobu: I don’t see any reason why this is built-in

cache[key] ||= initalization

class Hash
  def fetch_set(key)
    if self.key?(key)
      self[key]
    else
      if block_given?
        self[key] = yield
      else
        raise ...
      end
    end
  end
end

CACHE = {}
def memo_foo(*args)
  CACHE[args] ||= begin
    ... # mame doesn't like indent here (but this proposal is irrelevant)
  end
end

CACHE = {}
def memo_foo(*args)
  return CACHE[args] if CACHE.has_key?(args)
  ...
  return CACHE[args] = ret
end

cache.fetch_set(key) { initialization }

cache.fetch(key) { cache[key] = initialization }

# rejected
cache.fetch(key, set: true) { initialization }

store.fetch_or_set(key) { initialization }
store.init(key) { initialization }
store.cache(key) { initialization }
store.memoize(key) { initialization }

Discussion:

zsh % ./miniruby --dump=i -e'{}[k]||=v'
== disasm: #<ISeq:<main>@-e:1 (1,0)-(1,9)> (catch: FALSE)
putnil                                                           (   1)[Li]
newhash                                0
putself
opt_send_without_block                 <calldata!mid:k, argc:0, FCALL|VCALL|ARGS_SIMPLE>
dupn                                   2
opt_aref                               <calldata!mid:[], argc:1, ARGS_SIMPLE>
dup
branchif                               23
pop
putself
opt_send_without_block                 <calldata!mid:v, argc:0, FCALL|VCALL|ARGS_SIMPLE>
setn                                   3
opt_aset                               <calldata!mid:[]=, argc:2, ARGS_SIMPLE>
pop
leave
setn                                   3
adjuststack                            3
leave

Conclusion:

matz: The use case looks artificial. And I’m not fully satisfied the name fetch_or_set. I reject init cache memoise.

[Feature #17361] lambda(&block) does not warn with lazy proc allocation (znz)

optimization affects warnings

Preliminary discussion:

ko1: will do

block = proc {p 12}
lambda(&block) #=> warning
#=> warning: lambda without a literal block is deprecated; use the proc without lambda instead

it = foo{p 12}
it.call #=> 12

def foo &block
  lambda(&block) #=> no warning (bug? it should print a warning)
end

it = foo{p 12}
it.call #=> 12

Conclusion:

ko1: This is a bug, let me fix.

[Feature #17314] Method declarations (marcandre)

Have Module#protected, private, public accept a single Array argument of symbols (and also private_class_method/public_class_method for consistency)
Have Module#attr_reader, attr_writer, attr_accessor return an array of the method newly defined methods instead of nil
Have Module#alias_method return the symbol of the method defined instead of the receiver
Notes: All points are related but not dependant and are requested separately. Forwardable#def_delegators and ActiveSupport’s delegate already return array of symbols

Preliminary discussion:

private [:foo, :bar] # Allow this
attr_accessor :foo, :bar #=> [:foo, :foo=, :bar, :bar=] instead of nil

private attr_accessor :foo, :bar

$ ruby2.2 -w -e 'class X; private; attr :foo; end'
-e:1: warning: private attribute?
$ ruby2.3 -w -e 'class X; private; attr :foo; end'

nobu: how about private *attr_accessor :foo
mame: parentheses are required
nobu: What about alias?

Conclusion:

matz: no idea… okay, accepted

[Feature #17363] Timeouts (marcandre)

Add timeout parameter to Queue.pop, Ractor.receive, …
And/or Timeout::wake that would make these operations safer by raising only if thread is asleep (and presumably in a safe state)

Preliminary discussion:

ko1: I’ll add timeout to Ractor.receive, yield, take, and select. Will do by 3.0.
ko1: I have no opinion about Queue.pop
ko1: Timeout.wake should be discussed towards 3.1

Discussion:

ko1: I’d like to confirm: queue.pop(timeout: 0) always fails?
mame: Doesn’t queue.pop(timeout: 0) return a value if any value is ready to pop?
ko1: Unsure.
akr: The behavior must be explicitly written in rdoc
ko1: What’s happen when timeout? I thought that it returns nil
mame: doesn’t it raise an exception? Unless, it cannot distinguish nil’s meaning like the issue #17357
akr: raising an exception is slow but returning nil is ambiguous.
mame: Then, will exception: false be also wanted?
ko1: It looks difficult to decide it by Ruby 3.0

Conclusion:

ko1: will reply

[Feature #17357] `Queue#pop` should have a block form for clsoed queues (marcandre)

Adding block form similar to fetch ok?

Preliminary discussion:

mame: I’m unsure if this is really needed. We can easily work around, for example, by wrapping the value by an array. I’m not against, though.
ko1: no objection.

Discussion:

nobu: The same discussion was done when introducing Queue#close
naruse: Unsure about the use case

Conclusion:

naruse: postpone it to 3.1

[Feature #15975] Add `Array#pluck`/`Enumerable#pluck` method (connorshea)

It pulls values out of an array of hashes (e.g. array of user hashes, users.pluck(:username) gets all their usernames)
Please reconsider adding the method. I’ve added a comment to the issue with an example of when it’d be useful for me.
Is matz opposed to the method existing at all, or just the name?

Preliminary discussion:

mrkn: Use activesupport.

[{:foo => 1, :bar => 2}, {:foo => 3, :bar => 4}].pluck(:foo)   #=> [1, 3]
[{:foo => 1, :bar => 2}, {:foo => 3, :bar => 4}].map{_1[:foo]} #=> [1, 3]

Conclusion:

matz: agreed with rejection

[Feature #12650] Use UTF-8 encoding for ENV on Windows (larskanis)

See this with the next topic together

Preliminary discussion:

mame: already merged

Conclusion:

naruse: okay

[Feature #16604] Set default for Encoding.default_external to UTF-8 on Windows (larskanis)

Both have been postponed to ruby-3.0 years ago
default_external = UTF-8 is already de facto standard

Preliminary discussion:

mame: already merged

Conclusion:

naruse: okay

[Feature #17365] Adding `channel` to Ractor push api (marcandre)

yes/no?

Preliminary discussion:

ko1: I have counter proposal to add a pattarn-matching Ractor receive

Conclusion:

ko1: let’s discuss this in #17378

[Bug #15661] Disallow concurrent Dir.chdir with block (jonathanhefner)

This change is causing some issues in Rails (e.g. https://github.com/rails/rails/commit/ae5ecfe26c8 and https://github.com/rails/rails/issues/40756).
I have proposed an “escape hatch” mechanism in https://bugs.ruby-lang.org/issues/15661#note-15. Summary: wait until the end of the initial chdir block to check and raise error.
Can this be addressed for Ruby 3.0?

Preliminary discussion:

nobu: I think it is good to raise an exception only when it is called in a different thread
mame: I’m unsure if the escape hatch is okay. We cannot detect if pwd is unintentionally changed (especially, by a third-party library) in a block chdir.

Discussion:

nobu’s proposal: https://github.com/ruby/ruby/pull/3861/files

# Rakefile
task :foo do
  Dir.chdir("somedir")
end

# The implementation of Rakefile
def task(name)
  Dir.chdir("aaa") do
    yield
  end
end

# We need to change the code above to:
def task(name)
  pwd = Dir.pwd
  Dir.chdir("aaa")
  yield
ensure
  Dir.chdir(pwd)
end

# By this proposal
def task(name)
  Dir.chdir("aaa") do
    yield
    Dir.chdir("aaa")
  end
end

# The warning makes sense even in single-thread program
Dir.chdir("aaa") do
  third_party_method # may call Dir.chdir("bbb")
  # cwd is changed; it is unintentional
end

ko1: How about raising an exception if different thread calls Dir.chdir, and showing a warning if the same thread calls Dir.chdir
When Thread A is running Dir.chdir("aaa") do ... end
- Thread B attempts to run Dir.chdir("bbb") #=> an exception
- Thread B attempts to run Dir.chdir("bbb") { ... } #=> an exception
- Thread A attempts to run Dir.chdir("bbb") #=> a warning
- Thread A attempts to run Dir.chdir("bbb") { ... } #=> no warning

Dir.chdir("aaa") do
  Thread.new do
    Dir.chdir("bbb")    # exception
    Dir.chdir("bbb") {} # exception
  end
end

Dir.chdir("aaa") do
  Dir.chdir("bbb") # warning
end

Dir.chdir("aaa") do
  Dir.chdir("bbb") do # ok
  end
end

Conclusion:

matz: go ahead

DevelopersMeeting20201210Japan

DateTime and location

Next Date

Announce

About 3.0 timeframe

3.0 urgent tickets

[Feature #17371] Reintroduce expr in pat (ktsj)

[Bug #17369] Introduce non-blocking Process.wait, Kernel.system and related methods.

About naming

About hooks

About data structure

About marshalling

[Feature #17303] Make webrick to bundled gems or remove from stdlib (hsbt)

[Feature #17351] Deprecate Random::DEFAULT (eregon)

[Feature #17323] Ractor-local storage (marcandre)

[Feature #17378] Ractor#receive with filtering like other actor langauge (ko1)

report: rb_ext_ractor_safe(bool flag)

normal tickets

[Feature #16697] Hash.ruby2_keywords_hash?(value) should support any object (eregon)

[Feature #17055] Allow suppressing uninitialized instance variable and method redefined verbose mode warnings (jeremyevans0)

[Bug #17280] Dir.glob with FNM_DOTMATCH matches “..” and “.” and results in duplicated entries (jeremyevans0)

[Bug #17162] Dir[‘*/’] : stack smashing detected when listing big amount of directories (jeremyevans0)

[Bug #17218] Range#step sometimes behaves unexpectedly with Rational endpoints and increment (jeremyevans0)

[Bug #13663] String#upto doesn’t work as expected (jeremyevans0)

[Feature #17342] Add Hash#fetch_set (maxlap)

[Feature #17361] lambda(&block) does not warn with lazy proc allocation (znz)

[Feature #17314] Method declarations (marcandre)

[Feature #17363] Timeouts (marcandre)

[Feature #17357] Queue#pop should have a block form for clsoed queues (marcandre)

[Feature #15975] Add Array#pluck/Enumerable#pluck method (connorshea)

[Feature #12650] Use UTF-8 encoding for ENV on Windows (larskanis)

[Feature #16604] Set default for Encoding.default_external to UTF-8 on Windows (larskanis)

[Feature #17365] Adding channel to Ractor push api (marcandre)

[Bug #15661] Disallow concurrent Dir.chdir with block (jonathanhefner)

[Feature #17371] Reintroduce `expr in pat` (ktsj)

[Bug #17369] Introduce non-blocking `Process.wait`, `Kernel.system` and related methods.

[Bug #13663] `String#upto` doesn’t work as expected (jeremyevans0)

[Feature #17357] `Queue#pop` should have a block form for clsoed queues (marcandre)

[Feature #15975] Add `Array#pluck`/`Enumerable#pluck` method (connorshea)

[Feature #17365] Adding `channel` to Ractor push api (marcandre)