DevMeeting-2023-09-14

https://bugs.ruby-lang.org/issues/19858

DateTime and location

2023/09/14 (Thu) 13:00-17:00 JST @ Online

Next Date

Next: 2023/10/12 (Thu) 13:00-17:00 JST @ Online

Announce

About release timeframe

naruse: I want to have a preview release. Hold on, am packaging right now.

Ordinary tickets

[Feature #19075] Binary searching for the last element (kyanagi)

My latest pull request implements bsearch(target: :last) and bsearch_index(target: :last).
https://bugs.ruby-lang.org/issues/19075#note-6
argument name?: target: :last, find: :last, reverse: true, …
separate method or not?: bsearch_last, reverse_bsearch, rbsearch, …

Preliminary discussion:

ary = [1, 2, 3, 3, 3, 4]
ary.bsearch_index{|x|x>=3} #=> 2
ary.bsearch_index{|x|x>3}  #=> 5
ary.bsearch_index(last: true) {|x| x<3 } #=> 1

[1,2,3,4,5].bsearch{|x|x/2<=>1} #=> 3
[1,2,3,4,5].bsearch(last: true){|x|x/2<=>1} #=> 4

ary = ["A", "B", "C", "D", "E", "F"]
ary.bsearch_index {|x| x >= "A" } #=> 0
ary.bsearch_index {|x| x >= "D" } #=> 3
ary.bsearch_index {|x| x >= "G" } #=> nil

# proposed
ary.bsearch_last_index {|x| x < "D" } #=> 2

# If we implment the propsed one using current methods...
i = ary.bsearch_index {|x| !(x >= "D") }
if i == 0
  i = nil
elsif i
  i = i - 1
else
  i = ary.size - 1
end

ary.bsearch_index(target: :left) {|x| x >= "D" } #=> 2
ary.bsearch_index(search_last_false: true) {|x| x >= "D" } #=> 2

ary = ["A", "B", "C", "C", "C", "D", "E"]
ary.bsearch_index {|x| x <=> "C" }

Discussion:

akr: what does it mean by “last”?
mame: I was wondering that too.
mame: OP wants the “last index that satisfies the block” instead of the first one.
akr: what should be returned when there is no such thing that satisfies the block then? Not obvious to me.
mame: the proposed functionality is hard to implement by hand.
akr: that’s because bsearch itself is complicated by design.

mame: it’s hard to name the functionality.
akr: what about bsearch_last_found?
mame: the thing is not always “last found” I guess.

Conclusion:

matz: It looks difficult to design a good API for this use case. I want to reject it until a good API is proposed. I will reply

[Bug #19868] `Process::Status` methods for compatibility with `Fixnum` (nobu)

Looking at 54274b8c65a0981f1c69055a1513ba3c614dd675, noticed that pst_rshift contains shift by a negative integer, an unspecified behavior in C99.
I think an exception would be ok for such argument, also we can suggest alternative methods to encourage the transition.
https://github.com/ruby/ruby/pull/8392

Discussion:

deletion targets?
- ~~Process::Status#to_int~~
- Process::Status#>>
- Process::Status#&
nobu: We wrote them in 1.8 era.
shyouhei: the pull request is full of magic numbers…
akr: is it okay to delete them just because they are old?
nobu: they are not only old but also system-dependent. Very hard to use properly.

Conclusion:

matz: Let’s give it a try

[Bug #17146] Queue operations are allowed after it is frozen (jeremyevans0)

Last discussed at October 2020 developer meeting, no decision made.
@headius and I discussed this at RubyKaigi and agreed it would be best to raise FrozenError for push/pop for frozen queues.
Can we either decide this is a bug and fix it, or decide it is not and close it?

Preliminary discussion:

shyouhei: the previous discussion

Discussion:

ko1: not strange to raise FrozenError, but there are other cases when frozen things can be changed.
matz: for instance?
ko1: binding.freeze.eval("...") can change the local variable inside of it.
matz: can be an option to prohibit freezing a queue at all.
ko1: or prohibit operatins for frozen queue. In that case what happens if dequeueing a frozen-empty-queue is not obvious.
mame: the thread shall wait forever?
matz: that (non-obvoius-ness) is the reason why I’m for prohibiting freezing.
matz: queue being destructive is the nature of that class in general.
matz: I guess there are other methods that do not check frozen-ness in middle of that method.
ko1: like Array#map!?

ary = [1, 2, 3, 4, 5]

t = Thread.new do
  ary.map! {|x| sleep 1; x.to_s }
end

sleep 2
ary.freeze

p t.value # after 3 seconds
#=> can't modify frozen Array: ["1", "2", 3, 4, 5] (FrozenError)

mame: Queue being unable to get frozen sounds the easiest way to me.
shyouhei: NoMethodError?
mame: ENV.freeze raises TypeError

Conclusion:

matz: I think Queue#freeze does not make sense. I will reply

[Bug #17354] Module#const_source_location is misleading for constants awaiting autoload (jeremyevans0)

Last discussed at March 2021 developer meeting, no decision made.
I agree the method result is misleading, but as @mame states in note 10, current design is the most flexible.
Can we either decide this is a bug and fix it, or decide it is not and close it?

Preliminary discussion:

shyouhei: previous discussion

Discussion:

options:
- 1. as is
- 1. nil
- 1. load the file and return the location
- 1. an empty array
matz: I don’t like 2 and 4 either. 1 and 3 work for me. I think 1 is good enough because it is easy to load it actually be accessing the constant

Conclusion:

nobody objects for option #1.
matz: I will reply

[Bug #7964] Writing an ASCII-8BIT String to a StringIO created from a UTF-8 String (jeremyevans0)

Do we want StringIO to handle encoding?
Making it handle encoding makes it more like IO, fixing some existing compatibility issues but introducing new compatibility issues.

Preliminary discussion:

Discussion:

nobu: StringIO does not understand encoding options right now.
nobu: I have been dreaming of implementing this area…
mame: What should happen when a binary string is written for (theoretical) Unicode StringIO object?
usa: They want UndefinedConversionError, same as IO.
nobu: that’s sort of a hutsle.

Conclusion:

nobu: let me do this!

[Bug #9115] Logger traps all exceptions; breaks Timeout (jeremyevans0)

Can we fix this by supporting specific exception classes to reraise instead of swallow?
If not, is there an alternative approach that would be acceptable to fix this?

Discussion:

shoyouhei: Logger wants to continue writing at all costs vs Timeout wants to throw at all costs.
ko1: this shall be done by handle_interrupts.
mame: that will stop timeout to be raised.
ko1: logger do not consider asynchronous exception, it seems to me.

Conclusion:

there is no “fix” to this issue.
shyouhei: handle_interrupts is a better-than-nothing mitigation.
nobody is against inserting a handle_interrupts guard

Thread#native_thread_id is incorrectly cached across fork on Linux

[#19873]

ko1: seems nice.
naruse: will take a look.

Unicode line and paragraph separator are not stripped

[#19867]

by design? -> Yes

Update license phrases to SPDX BSD-2-Clause

[#19860]

usa: why the change?
hsbt: “BSD-2-Clause” by SPDX is different from current our BSDL. Should we update it?
Matz: There is no opinion.
naruse: SPDX-listed licenses could be machine-detected
naruse: but Ruby’s license is not machine-detectable at the beginning (we are combining lots of differently licensed source codes)
hsbt: ok

Start & Finish, Begin & End

[#19859]

matz: start_with? is an influence from java.
mame: Python and JS also uses startsWith/endsWith
rejected

Get thread creation time

[#19850]

usa: understand the use case.
usa: but the thread could be reused using thread pools
ko1: maybe API to reset the time?
usa: would better be done by users.
naruse: modify the thread’s name accordingly.
mame: or assign an instance variable.

Requiring file with autoload results in confusing error if file doesn’t exist

[#19849]

usa: we can show autolaod line in LoadError message.

Need a method to check if two ranges overlap

[#19839]

matz: accepted

p (1..2).overlap?(2...2)  #=> true -> false?
p (2..2).overlap?(2...2)  #=> true -> false?
p (2...2).overlap?(2...2) #=> true -> false?
p (1..2).overlap?(2..-1)  #=> true -> false?
p (1..2).overlap?(3..-1)  #=> false
p (4..6).overlap?((2..8) % 3) #=> TypeError (okay?)
p (3..4).overlap?((2..8) % 3) #=> TypeError (okay?)

Method#destructive?, UnboundMethod#destructive?

[#19832]

matz: I will reject

Allow `Array#transpose` to take an optional size argument

[#19830]

# basic expectation
ary.transpose.transpose == ary

# but is not true
[[], []].transpose.transpose #=> []

[[], []].transpose.transpose[0] #=> expected: [], actual: nil

# because the first transpose returns an empty array
[[], []].transpose #=> []
# (2, 0)               (0, 2) -> (0, 0) dimension

# explicitly specify the dimension
[[], []].transpose.transpose(2) #=> [[], []]

matz: I cannot determine if this is needed
knu: .fill(0, 2) { [] }

Add Enumerable#uniq_map, Enumerable::Lazy#uniq_map, Array#uniq_map and Array#uniq_map!

[#19787]

Remove tailcall_optimization support

[#19780]

Introduce defp keyword for defining overloadable, pattern matched methods

[#19764]

URI::HTTP.build does not accept a host of `_gateway`, but `URI.parse` will.

[#19756]

URI::HTTP.build(host: "gateway")    #=> #<URI::HTTP http://gateway>
URI::Generic.build(host: "gateway") #=> #<URI::Generic //gateway>

$ irb -ruri
irb(main):001> URI::DEFAULT_PARSER = URI::RFC3986_PARSER
(irb):1: warning: already initialized constant URI::DEFAULT_PARSER
/home/mame/.rbenv/versions/3.3.0-dev/lib/ruby/3.3.0+0/uri/common.rb:24: warning: previous definition of DEFAULT_PARSER was here
=> #<URI::RFC3986_Parser:0x00007f04fdbc3488>
irb(main):002> URI::HTTP.build(host: "_gateway")
=> #<URI::HTTP http://_gateway>

If there are real people who are in a trouble of this, we may think about it later

Add support for UUID version 7

[#19735]

UUIDv7 contains of not just a random value, but also a timestamp. Is it okay to provide it in random/formatter? It will not be idempotent

Warning for non-linear Regexps

[#19720]

To consider both compatibility and security…

定義: 「リニアでない正規表現」を「make_now_justさんので動かない正規表現」正規表現とする

back reference
- check /([”’])[a-z]*\1/ on https://makenowjust-labs.github.io/recheck/playground/
sub expression call
普通に作った正規表現
- -wなし: 警告でない
- -W:may-nonlinear-regexpあり: 警告でる
looseモードの正規表現 /(x)\1/d
- -wなし: 警告でない
- -W:may-nonlinear-regexpあり: 警告でない
Matz: 正規表現の素晴らしさをspoilするような機能は入れたくない
- -W:may-nonlinear-regexpはギリギリ許容

DevMeeting-2023-09-14

DateTime and location

Next Date

Announce

About release timeframe

Ordinary tickets

[Feature #19075] Binary searching for the last element (kyanagi)

[Bug #19868] Process::Status methods for compatibility with Fixnum (nobu)

[Bug #17146] Queue operations are allowed after it is frozen (jeremyevans0)

[Bug #17354] Module#const_source_location is misleading for constants awaiting autoload (jeremyevans0)

[Bug #7964] Writing an ASCII-8BIT String to a StringIO created from a UTF-8 String (jeremyevans0)

[Bug #9115] Logger traps all exceptions; breaks Timeout (jeremyevans0)

Thread#native_thread_id is incorrectly cached across fork on Linux

Unicode line and paragraph separator are not stripped

Update license phrases to SPDX BSD-2-Clause

Start & Finish, Begin & End

Get thread creation time

Requiring file with autoload results in confusing error if file doesn’t exist

Need a method to check if two ranges overlap

Method#destructive?, UnboundMethod#destructive?

Allow Array#transpose to take an optional size argument

Add Enumerable#uniq_map, Enumerable::Lazy#uniq_map, Array#uniq_map and Array#uniq_map!

Remove tailcall_optimization support

Introduce defp keyword for defining overloadable, pattern matched methods

URI::HTTP.build does not accept a host of _gateway, but URI.parse will.

Add support for UUID version 7

Warning for non-linear Regexps

[Bug #19868] `Process::Status` methods for compatibility with `Fixnum` (nobu)

Allow `Array#transpose` to take an optional size argument

URI::HTTP.build does not accept a host of `_gateway`, but `URI.parse` will.