Sep 2024 Meeting

calendar_today September 05, 2024 confirmation_number 16 tickets discussed Source English

DevMeeting-2024-09-05

https://bugs.ruby-lang.org/issues/20660

DateTime and location

  • 2024/09/05 (Thu) 13:00-17:00 JST @ Online

Next Date

  • 2024/10/03 (Thu) 13:00-17:00 JST @ Online
    • matz: 10/10 is NG to me

Announce

About release timeframe

Misc

r-l.o email

:@shyouhei: I still cannot send/receive using my ruby-lang.org email address. Can you help?

Ordinary tickets

[Feature #20627] require on Ractor should run on the main Ractor

Discussion

  • ko1: The idea is already approved, what remains is the naming.
  • ko1: Ractor.require :arrow_right: Ractor._require done.
  • ko1: Also I need Ractor._activated which runs at most once per process lifetime.
  • nobu: I’m not quite sure if the proposed implementation works.

Conclusion

  • matz: accepted, as far as the bug mentioned by @nobu is fixed.

[Feature #15554] warn/error passing a block to a method which never use a block (Re: strict mode)

Discussion

  • ko1: I tried this and saw lots (thousands) of false positives.
  • ko1: we need strict / non-strict modes for this warning.
  • ko1: how we do that?
  • nobu: NG for environment variables only.
  • shyouhei: library authors might want to suppress this warning for their own gems.
  • akr: that would not happen when relax mode is the default.
  • matz: maybe in a future we want code annotations… but not today.
  • mame: How about issue warning only outside of gems?
  • nobu: One can filter out warnings inside of gems using Warning.warn.

Conclusion

  • no conclusion yet, continue discussion.

[Feature #20594] A new String method to append bytes while preserving encoding (byroot)

  • Continuation from previous meeting
  • Alan proposed the method to be named append_as_bytes, to be variadic and accept both String and Integer (so like the original proposal).
  • For integer the behavior would be the same as setbyte, so mask 0xff.
  • Early implementation prototype show a ~13% gain on our benchmark.

Preliminary discussion:

# buf <<
#   255 << title.bytesize << title <<
#   255 << body.bytesize << body

buf = ""
buf.append_as_bytes(255, title.bytesize, title, 255, body.bytesize, body)
  • buf.append_as_bytes(256) should write "\x00", not raise an exception
    • as like buf.~~setbyte~~(256)

Discussion:

  • matz: nobody is against the API?
  • ko1: maybe “as” is the new terminology.
  • matz: sounds better than byteappend to me.

Conclusion:

  • matz: accepted

[Feature #20684] Add optimized instructions for frozen literal Hash and Array (byroot)

  • Allow to avoid having to define EMPTY_HASH = {}.freeze or EMPTY_ARRAY.freeze constant in hotspots (see code samples in the issue).
  • Also give some minor gains when defining frozen constants.
  • Not a big optimization, but a very small patch and help usability.

Discussion:

  • matz: what prevents this from being approved?
  • mame: Optimisation of String#freeze eventually introduced frozen-string-literal. This can also?
  • akr: I don’t think magic comments work for empty arrays etc.
  • ko1: This proposal changes the identity of frozen empty hash.
  • nobu: It seems the proposed patch has a problem (might freeze non-empty arrays).
    • [1].freeze is also optimized (but not dedup’ed?)

Conclusion:

  • matz: I see no problem. [1].freeze is cruby’s implementation detail that should not be a stable specification of the language.

discussion

  • mame: my investigation is that because of vwa this reduces the frequency of full gc.
  • ko1: my theory is vwa might not handle situations when only one heap is used very well.
  • shyouhei: I’m not entirely sure what is going on here. We need better observability.

conclusion

  • matz: this is a bug that should be fixed.

[Bug #20675] Parse error with required kwargs and omitted parens (jeremyevans0)

  • Is consistency with method parameters worth breaking compatibility in this case?

Discussion:

  • mame: what is your expectation?
  • matz: would be simpler to make them continuous, but backward compatibility is a concern.

Conclusion:

  • matz: I replied

[Feature #20309] Bundled gems for Ruby 3.5 (hsbt)

  • Discuss about benchmark,irb and reline

Discussion:

  • shyouhei: what happens for binding.irb?
  • hsbt: we need to take care. But I guess it’s possible to keep the current behaviour. Let me try.
# t.rb
gem 'debug'
require 'debug'

# Gemfile is empty

#=>
$ bundle exec ruby t.rb
/home/ko1/ruby/install/trunk/lib/ruby/3.4.0+0/bundler/rubygems_integration.rb:253:in 'block (2 levels) in Kernel#replace_gem': debug is not part of the bundle. Add it to your Gemfile. (Gem::LoadError)
        from t.rb:1:in '<main>'

ko1: add a method which load newest gem on “bundle exec” context in bundler or rubygems?

Other libraries:

  • mame: default to standard libs for some libs?
  • hsbt: Singleton or tempfile? Discuss next year.
  • hsbt: erb has value to be default gem which can be updated.
``` ext/date/date.gemspec ext/digest/digest.gemspec ext/etc/etc.gemspec ext/fcntl/fcntl.gemspec ext/fiddle/fiddle.gemspec ext/json/json.gemspec ext/openssl/openssl.gemspec ext/pathname/pathname.gemspec ext/psych/psych.gemspec ext/stringio/stringio.gemspec ext/strscan/strscan.gemspec ext/win32ole/win32ole.gemspec ext/zlib/zlib.gemspec lib/English.gemspec lib/bundler/bundler.gemspec lib/cgi/cgi.gemspec lib/delegate.gemspec lib/did_you_mean/did_you_mean.gemspec lib/erb.gemspec lib/error_highlight/error_highlight.gemspec lib/fileutils.gemspec lib/find.gemspec lib/forwardable/forwardable.gemspec lib/ipaddr.gemspec lib/logger/logger.gemspec lib/net/net-protocol.gemspec lib/open-uri.gemspec lib/open3/open3.gemspec lib/optparse/optparse.gemspec lib/ostruct.gemspec lib/pp.gemspec lib/prettyprint.gemspec lib/prism/prism.gemspec lib/pstore.gemspec lib/rdoc/rdoc.gemspec lib/readline.gemspec lib/resolv.gemspec lib/ruby2_keywords.gemspec lib/securerandom.gemspec lib/set/set.gemspec lib/shellwords.gemspec lib/singleton.gemspec lib/syntax_suggest/syntax_suggest.gemspec lib/tempfile.gemspec lib/time.gemspec lib/timeout.gemspec lib/tmpdir.gemspec lib/tsort.gemspec lib/un.gemspec lib/uri/uri.gemspec lib/weakref.gemspec lib/yaml/yaml.gemspec ```

Conclusion:

*

[Bug #19744] Namespace on read (mame)

current situation

  • tagomoris: currently I have one remaining segv. Other than that test-all is passing.
  • mame: is make test-bundled-gems green?
  • hsbt: make check is recommended for your daily life.
  • tagomoris: I hope I’m going to issue a pull request this month.

timeline

  • matz: It’s September. When do we merge this?
  • shyouhei: someday until 12/24?
  • mame: I NEED @yahonda’s rails test long before holidays.
  • shyouhei: Maybe we want to push the merge button as soon as possible.
  • mame: Do we merge this when something breaks? Maybe yjit etc.?
  • matz: Okay for namespace not interacts with yjit, as far as this is experimental.

refinements

  • tagomoris: There are three situations.
    • builtin: Where classes like e.g. String reside. These classes shall be visible from everywhere (otherwise you cannot create a string from inside a namespace which is surreal.)
    • main: Where classes like e.g. Gem reside. This is the namespace that is implicitly created when the process boots up.
    • other userland namespaces that are created on-demand.
  • tagomoris: in my current implementation, main namespace’s Object class includes a module that interacts with gem_original_requrire. However builtin namespace cannot use this technique (no gem). So we use refinements there.
  • mame: bootsnap defines require in main. Then require inside of builtin namespace should use it…?
  • hsbt: in case of zeitwerk, zeitwerk’s require uses bootsnap’s require, which uses ruby’s original require.
  • mame: the problem is that when a program has require "foo", what happens is not obvious from the interpreter’s POV.
  • tagomoris: I don’t think requires in builtin namespaces need bootsnap.

detection of namespace

  • ko1: how do your requrie detect the destination namespace?
  • tagomoris: I thought I can scan CREF.
  • ko1: maybe you can scan the stack?
  • ko1: or maintain a thread local variable.
  • mame: when that happens?
  • ko1: inside of a require.
  • mame: the “require” here is not obvious because it can be redefined.

[Bug #20693] Dir.tmpdir should perform a real access check before warning about writability (kjtsanaktsidis)

  • A minor paper-cut, but I ran into this while running Ruby’s tests in an unprivileged Docker container
  • File::Stat#writable? is really just guessing; to really know if a directory is writable, you need to ask the OS (through a call to File.writable?)
  • There’s a broader question about whether we should deprecate File::Stat#writable? and File::Stat#readable? altogether, since they can’t possibly be correct in 100% of cases. But I’m happy just to fix this one case in Dir.tmpdir for now.

Discussion:

  • mame: in short KJ wants to use File.writable? instead of File::Stat
  • nobu: not against it.
  • nobu: The world-writable check below could also be affected by ACLs.
  • mame: true, but we cannot say if a world-writable directory is in fact protected by ACLs or not.

Conclusion:

  • akr: I want a comment here why this change is necessary. Otherwise LGTM.

[Feature #20707] Move Time#xmlschema into core (byroot)

Discussion:

  • akr: RSS has another implementation of ISO8601: w3cdtf method.
  • mame: Time#xmlschema is not everything that ISO8601 can represent.
  • akr: Yes. ISO8601 permits many variants.
  • nobu: Time.new now accepts this string so generation could also be in core.
  • matz: I don’t really like the “xmlschema” naming though.

Conclusion:

  • matz: Accepted
  • matz: Time#iso8601 is also accepted to move it to the core

[Feature #20673] Enable native SOCKS support by default (znz)

Discussion:

  • matz: This pull request builds, but mandates a socks runtime library.
  • akr: -1 for additional runtime dependency.

[Bug #20708] EINTR while opening fifo isn’t retried

$ rm -f /tmp/badger && mkfifo /tmp/badger && ./ruby --disable-gems -we '
trap(:USR1) {}
Thread.new { sleep 1; Process.kill(:USR1, $$) }
IO.read("/tmp/badger")
'
-e:4:in 'IO.read': Interrupted system call @ rb_sysopen - /tmp/badger (Errno::EINTR)
        from -e:4:in '<main>'

Conclusion:

  • nobu fixed already.

[Feature #20705] Should “0.E-9” be a valid float value?

Discussion:

  • mame: 0.E-9 is (0.).E(-9).
  • nobu: The problem is Float("0.E-9")

```ruby:Current behavior Float(“0.E-9”) #=> invalid value for Float(): “0.E-9” Float(“1.”) #=> invalid value for Float(): “1.” “0.E-9”.to_f #=> 0.0 “1.E-9”.to_f #=> 1.0 (expected?: 1.0e-09) “1.”.to_f #=> 1.0 Float(“0x1p+0”) #=> 1.0 “0x1p+0”.to_f #=> 0.0


```ruby:matz
Float("1.E-9") #=> 1.0e-09
Float("1.")    #=> 1.0

[Feature #20703] Alias StringIO#string to StringIO#to_s/to_str

require "stringio"
sio = StringIO.new('my string')

# current behavior
p sio    #=> #<StringIO:0x00007ff905a62e70>
puts sio #=> #<StringIO:0x00007ff905a62e70>
  • matz: to_str must not be changed
  • matz: it is possible to improve StringIO#inspect as showing its size, but it is a different issue
  • nobu: I will reject

[Feature #20702] Add Array#fetch_values

irb(main):001> ENV.methods - {}.methods
=> []
irb(main):002> {}.methods - ENV.methods
=>
[:compare_by_identity?,
 :compare_by_identity,
 :deconstruct_keys,
 :<=,
 :>=,
 :dig,
 :<,
 :>,
 :to_proc,
 :flatten,
 :compact!,
 :default=,
 :default_proc,
 :default_proc=,
 :default,
 :transform_keys,
 :transform_keys!,
 :transform_values,
 :transform_values!,
 :fetch_values,
 :merge]

h = { a: 1, b: 2, c: 3 }
h.fetch_values(:b, :c) #=> [2, 3]
h.fetch_values(:b, :d) #=> IndexError
h.fetch_values(:b, :d){ 42 } # => [2, 42]

ENV.fetch_values # no method

a = [0, 1, 2]
a.fetch_values(0, 2) # => [0, 3]
a.fetch_values(0, 4) # => IndexError
a.fetch_values(0, 4){ 42 } # => [0, 42]
  • matz: Array#fetch_values is accepted
  • matz: ENV.fetch_values is also accepted

[Bug #20690] URI.encode_www_form_component method escapes tilde when it’s not supposed to

  • knu: The method does not strictly implement a specific RFC based behavior. It is provided as is and changing its behavior would likely break existing use cases.
  • naruse: WHATWG URL has the latest definition, which excludes the tilde from the characters that should be percent-encoded.

https://github.com/ruby/uri/pull/117

https://github.com/ruby/ruby/pull/10864 to @akr