DevelopersMeeting20160316Japan

Date: 2016/03/16 14:00- JST

Attendees: Matz, ko1, nobu, akr, naruse, mrkn, shyouhei, sorah, martin, hsbt

Next meeting

2016/04/13 13:00 @ Money Forward

[#10098] Timing-safe string comparison for OpenSSL::HMAC (naruse)

• (n0kada) name not fixed.
• Rack implements this in pure-ruby

• but sounds not a wise idea to do this in Ruby.

• It might be a good idea to use libc-provided one, if any.
• Should this be a String’s method?
• Why not gem?

• This kind of things are difficult to properly implement.  Worth providing by default.

• Does OpenSSL has this feature?

• Yes.  There is CRYPTO_memcmp.
• Why not just export this to ruby?

• naruse: How about OpenSSL.foo_bar() ?

• OpenSSL.memcmp() ?

• akr: Does it consider encoding?

• Naruse: Shouldn’t. It’ll be useful if it compares each other as a binary.

• OpenSSL.memcmp(str1, str2)

• The name means

• “OpenSSL”: secure (timing safe)
• “memcmp: binary comparison

• Return value?

• true/false.  -1/0/1 makes almost no sense for us.

[Feature #12125] Proposal: Shorthand operator for Object#method (hsbt)

• (Matz) I want this. But all proposed syntax so far don’t feel right.

• Proposed: -> .> .<...> .:

• Approaches:

• Shorthand for making Procs
• Shorthand for Object#method

• Matz: #method is a normal method that can be overridden. This kind of feature should be syntactic.
• Still waiting for a new syntax.

Onigmo maintenance (duerst)

• (Dürst) Who is the maintainer of onigumo?
• K-takata san (not that active though)
• (Naruse) Onigmo is on github, should send PR as usual when needed.

github.com/k-takata/Onigmo

upper/lower case conversion (duerst)

• (Dürst) under development.  Works most part.

• Want to make UTF-8 conversion by default, and explicit parameter passing for ASCII-conversion.

• Titlecase and String#swapcase

• In unicode there is a special case called titlecase cf http://unicode.org/faq/casemap_charprop.html#4
• How should swapcase behave?
• Or, how should swapcase behave in general?
• Other languages:

• Perl uses tr
• Python has swapcase.
• Also javascript

• Matz: maybe it’s from Python.

[Bug #11844] Please update unicode-licensed files (license issue) (shyouhei)

• This table was from NetBSD.
• The table is a data, not an “expression of author” that a copyright should apply.
• Is unicode’s license GPL compatible?
• http://www.unicode.org/Public/MAPPINGS/OBSOLETE/EASTASIA/JIS/JIS0212.TXT
• It seems unicode.org has version 2 of those files.
• Lets update our copy to the latest ones.

[Bug #11816] Partial safe navigation operator (shyouhei)

• (shyouhei) current status?

• No concrete answer

• (n0kada) there can be a flaw for e.g. binary operator, foo&.bar + 1 ?

• (a&.b.c.d.e...).foo -> foo should be called always.
• So it is safer than the opposite.
• Edge case : a&.b.c += i

• Other languages:

• groovy : no chain
• Coffee : it does
• Swift : it does
• C# : it does

[Feature #12142] Hash tables with open addressing (in particular, question about 32/64-bit indices) (duerst)

• 4G entry hash is rare, but not an unrealistic thing in long term.
• It is definitely a good thing to provide a fast alternative for 99% use case, however.
• Any ideas?

• 32bit by default?

• Is 4G entry hash a real thing today?  Can we wait for this a few more years?
• 4G entry requires 6word * 4G = 100GB

• Compile-time switch using configure?

• Not a wise idea because a user cannot distinguish which version of ruby they are running.

• Switch 16bit->32bit->64bit smoothly?

[Feature #12020] Documenting Ruby memory model (shyouhei)

• Koichi is going to dump his opinion to reply the thread.
• (Matz) I don’t want to bother such in-detail memory

[Bug #12039] Fixnum#infinite?/Bignum#infinite or Numeric#infinte, consistent with Float#infinite? and BigDecimal#infinite? (shyouhei)

• For about JSON, #finite? (not #infinite?) is much better because JSON also has to reject NaNs.
• (Matz) positive.  Float <-> Integer polymorphism is a good thing.
• Add both #finite? and #infinite?

[Feature #12005] Unify Fixnum and Bignum into Integer (shyouhei)

• Nobody is negative.
• Mrkn is assigned for this
• (n0kada) coerce can be a problem.
• (usa): Is the internal implementation changed?

1. Current Fixnum and Bignum C implementation move under Integer

• (usa): Will C API be changed?

1. Expected to keep compatibility as far as we can.
2. For example rb_fix_hoge, and LONG2FIX.

• Proposed migration process

1. Move every method from Fixnum/Bignum to Integer
   This solves Fixnum/Bignum rdoc duplication problem.
2. Delete bignum; ::Bignum = ::Integer

Gem search (duerst)

• Gem codesearch spec
• Codesearch is done using rubygems-mirror
• 500GB storage, memory unknown

[ruby-core:74355] Ruby+OMR: how should we proceed?

• Thanks a lot for contacting us.
• Normal pull-request is the ideal way.
• We have to look at actual code to judge details.

[Feature #12026] Support warning filters (jeremyevans0)

• This is related to https://bugs.ruby-lang.org/issues/11588
• (Matz) I thought structured warning was overkill.
• (naruse) how about squashing all seen warning strings.
• (ko1) regular expression seems too shortcoming.
• (akr) It’s not a bad idea to provide a primitive to do something on warnings.

[Feature #12092] Allow Object#clone to yield cloned object before freezing (jeremyevans0)

• Use case?
• There are two ways to create a cloned object, namely #clone and #dup.  The difference is singleton classes, and frozenness.
• (n0kada) why not avoid singleton method and use dup->extend->freeze maneuver?

• That way you have to manage all and every calls to extend

[Feature #12172] Array#max and Array#min (mame)

• Nobody is against Array#max
• (shyouhei): Why only opt_newarray_max?

• (mame) When Array include another module which redefines max/min, current implementation can’t detect it.
• (mame) opt_newarray_max is only for literal.

• Nobody except ko1 is against opt_newarray_max

• Koichi afraids that increase VM complexity and other people want to add other methods.
• However, [...].min or .max are used CPU intensive cases. So that it is acceptable. If other techniques are invented, then Koichi will remove them.

[Feature #9969] Add File.empty? as alias to File.zero? (shyouhei)

• (akr) File.zero? is too bad in naming.
• Nobody is against it.

[Bug #12121]異なる名前空間にある同名の定数により Module.constants の結果の並びが変わる (shyouhei)

• The reason behind this is we introduced id_table, which does not preserve order.
• We did not, and do not want to, warrant such thing.
• This is a doc issue.

[Bug #12112] Resolv.getname with IPv6 noop (shyouhei)

• Assign to akr

[Bug #12162] OpenSSL::PKCS7 seems to create broken objects (nested asn.1 error) (shyouhei)

• Assign to openssl

[Bug #12091] Freezing a SortedSet breaks Enumerable (shyouhei)

• Assign to knu

[Bug #12126] [PATCH] openssl: accept moving write buffer for write_nonblock (shyouhei)

• Seems ok

[Feature #10617] Change multiple assignment in conditional from parse error to warning (shyouhei)

• Koichi: i want to make it “void” expression for performance.

[Feature #12094] parameterized property assignment: o.prop(arg) = 1 (shyouhei)

• Nobu is going to update the current status

[Feature #12096] New notation for instance variables and class variables (shyouhei)

• This is metaprogramming.  And having a metaprogramming in syntax is ``wrong’’.
• Being able to write identical variable in different form is problematic (both for users and editors)

[Bug #12104] Procs keyword arguments affect value of previous argument (shyouhei)

• Ko1 will update his current opinion.

[Bug #12106] Behavior of double splatting of hashes with non symbol key is different according to splatted hash position (shyouhei)

• Assign to nobu.

[Feature #11997] A method to read a file with interpolations (shyouhei)

• Nobu thinks this is not that simple to do.  He will add comments about it.

[Feature #12116] Fixnum#divmod, Bignum#divmod with multiple arguments (shyouhei)

Good example.

[Feature #12119] next_prime for lib/prime.rb (shyouhei)

• Assign to yugui.

[Feature #12134] Comparison between true and false (shyouhei)

• Use partition method.

[Feature #11547] remove top-level constant lookup (shyouhei)

• Matz issue.

[Misc #12124] Use Automake (shyouhei)

• (naruse) common.mk is also used by nmake.
• (naruse) Current patch fails CI.
• Generally ok, but it must works with nmake.

[Bug #12139] return OpenSSL::Random.random_bytes(n) call takes to long. OpenSSL:: bug on windows. (naruse)

• (naruse) How about using Random.naw_seed on Windows
• (akr) ok while it doesn’t cost too much entropy.