May 2024 - ruby-core - ml.ruby-lang.org

[ruby-core:117497] [Ruby master Bug#20424] ZLib::GZipReader always double allocates strings when passed outbuf, significantly increasing memory usage
by martinemde (Martin Emde) 14 May '24

14 May '24

Issue #20424 has been reported by martinemde (Martin Emde). ---------------------------------------- Bug #20424: ZLib::GZipReader always double allocates strings when passed outbuf, significantly increasing memory usage https://bugs.ruby-lang.org/issues/20424 * Author: martinemde (Martin Emde) * Status: Open * ruby -v: ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] * Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN ---------------------------------------- In trying to improve the memory performance during the install of rubygems, we previously found a bug in `eof?`. Further investigation into the memory usage during the fix for this bug found wasteful allocating of strings in readpartial and read. In ZLib, when reading with readpartial or read, a new string is always created for the bytes read from the buffer. The current approach allocates a string no matter if there is an outbuf passed. ``` # vastly simplified psuedo implementation def readpartial(len, dst=nil) if (buffer.empty?) buffer = gzipfile.readpartial(len, dst) # adds inflated bytes into dst if passed end dst = allocate_new_string(len) # make a new string for the destination dst << buffer.read(len) # read from the buffer into the destination end ``` The result is that readpartial always allocated at least double the bytes necessary. Samuel Giddins submitted, and I have tested and reviewed, [a pull request, zlib#61](https://github.com/ruby/zlib/pull/61) that resolves the issue and vastly improves the memory usage and increases the speed of GZipReader by avoiding excess memcpy and rb_str_new calls that were wasted. This PR also adds an outbuf to GZipReader#read for improvement memory management, very similar to [IO#read](https://ruby-doc.org/core-2.5.1/IO.html#method-i-read) We appreciate your attention to this performance improvement. We believe it will further improve the performance of rubygems gem installs. -- https://bugs.ruby-lang.org/

1 1

[ruby-core:117880] [Ruby master Feature#6648] Provide a standard API for retrieving all command-line flags passed to Ruby
by Dan0042 (Daniel DeLorme) 14 May '24

14 May '24

Issue #6648 has been updated by Dan0042 (Daniel DeLorme). nobu (Nobuyoshi Nakada) wrote in #note-15: > `RbConfig` is for build time informations, and does not look a right place for runtime informations. Isn't it ok to relax the semantics a little bit? RbConfig seems to me the most logical place for "ruby configuration", both run time and build time. But this does bring the excellent point that `RbConfig.ruby` is not necessarily the location of the ruby interpreter as I previously thought: ``` $ ruby -e 'p RbConfig.ruby' "/opt/ruby/3.2/bin/ruby" $ cp /opt/ruby/3.2/bin/ruby rubyyyy $ ./rubyyyy -e 'p RbConfig.ruby' "/opt/ruby/3.2/bin/ruby" ``` So it's not quite suitable for re-executing. So we could either * change `RbConfig.ruby` to be the current ruby interpreter (because TBH I'm not sure what's the use of this current `RbConfig.ruby`) * add a new method like `RbConfig.ruby_executable` * use a different namespace like `Process.ruby` and `Process.ruby_args` --- headius (Charles Nutter) wrote in #note-16: > if there are quoted arguments on the command line they might have to be re-quoted by the user if they are passed through another shell to launch. Wouldn't you normally use Shellwords for this? The original quoting is not available to ruby anyway. ---------------------------------------- Feature #6648: Provide a standard API for retrieving all command-line flags passed to Ruby https://bugs.ruby-lang.org/issues/6648#change-108292 * Author: headius (Charles Nutter) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- Currently there are no standard mechanisms to get the flags passed to the currently running Ruby implementation. The available mechanisms are not ideal: * Scanning globals and hoping they have not been tweaked to new settings * Using external wrappers to launch Ruby * ??? Inability to get the full set of command-line flags, including flags passed to the VM itself (and probably VM-specific) makes it impossible to launch subprocess Ruby instances with the same settings. A real world example of this is "((%bundle exec%))" when called with a command line that sets various flags, a la ((%jruby -Xsome.vm.setting --1.9 -S bundle exec%)). None of these flags can propagate to the subprocess, so odd behaviors result. The only option is to put the flags into an env var (((|JRUBY_OPTS|)) or ((|RUBYOPT|))) but this breaks the flow of calling a simple command line. JRuby provides mechanisms to get all its command line options, but they require calling Java APIs from Ruby's API set. Rubinius provides its own API for accessing comand-line options, but I do not know if it includes VM-level flags as well as standard Ruby flags. I know there is a (({RubyVM})) namespace in the 2.0 line. If that namespace is intended to be general-purpose for VM-level features, it would be a good host for this API. Something like... ``` class << RubyVM def vm_args; end # returns array of command line args *not* passed to the target script def script; end # returns the script being executed...though this overlaps with $0 def script_args; end # returns args passed to the script...though this overlaps with ARGV, but that is perhaps warranted since ARGV can be modified (i.e. you probably want the original args) end ``` -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117879] [Ruby master Feature#15554] warn/error passing a block to a method which never use a block
by Edwing123 (Edwin Garcia) 14 May '24

14 May '24

Issue #15554 has been updated by Edwing123 (Edwin Garcia). `--warn-unused-block`/`:warn_unused_block`? ---------------------------------------- Feature #15554: warn/error passing a block to a method which never use a block https://bugs.ruby-lang.org/issues/15554#change-108291 * Author: ko1 (Koichi Sasada) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- # 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` doesn`t warn warning even if this method doesn't use a block. The rule (3) can pass blocks easily and there are many methods don`t 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 (commit: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/… 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-pat… 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#… 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-pat… 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? -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117494] [Ruby master Bug#20421] String#index and String#byteindex don't clear `$~` when offset > size (or bytesize)
by andrykonchin (Andrew Konchin) 14 May '24

14 May '24

Issue #20421 has been reported by andrykonchin (Andrew Konchin). ---------------------------------------- Bug #20421: String#index and String#byteindex don't clear `$~` when offset > size (or bytesize) https://bugs.ruby-lang.org/issues/20421 * Author: andrykonchin (Andrew Konchin) * Status: Open * ruby -v: 3.2.2 * Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN ---------------------------------------- When String#index and String#byteindex are called with Regexp and offset they clear `$~` (that contains result of the last Regexp matching) when offset < -size. But they don't clear `$~` when offset > size what seems not consistent: ```ruby "a".index(/a/) p $~ # => #<MatchData "a"> string = "blablabla" string.index(/bla/, string.length + 1) p $~ # => #<MatchData "a"> ``` -- https://bugs.ruby-lang.org/

5 6

[ruby-core:117875] [Ruby master Feature#6648] Provide a standard API for retrieving all command-line flags passed to Ruby
by headius (Charles Nutter) 14 May '24

14 May '24

Issue #6648 has been updated by headius (Charles Nutter). Note that for this to be most effective it would be the arguments unprocessed as they appear on the command line, but that may not be possible to do if the shell removes quoting. I don't think that should be a reason not to implement this issue, but if there are quoted arguments on the command line they might have to be re-quoted by the user if they are passed through another shell to launch. They should work fine if passed as direct arguments to `spawn`. ---------------------------------------- Feature #6648: Provide a standard API for retrieving all command-line flags passed to Ruby https://bugs.ruby-lang.org/issues/6648#change-108287 * Author: headius (Charles Nutter) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- Currently there are no standard mechanisms to get the flags passed to the currently running Ruby implementation. The available mechanisms are not ideal: * Scanning globals and hoping they have not been tweaked to new settings * Using external wrappers to launch Ruby * ??? Inability to get the full set of command-line flags, including flags passed to the VM itself (and probably VM-specific) makes it impossible to launch subprocess Ruby instances with the same settings. A real world example of this is "((%bundle exec%))" when called with a command line that sets various flags, a la ((%jruby -Xsome.vm.setting --1.9 -S bundle exec%)). None of these flags can propagate to the subprocess, so odd behaviors result. The only option is to put the flags into an env var (((|JRUBY_OPTS|)) or ((|RUBYOPT|))) but this breaks the flow of calling a simple command line. JRuby provides mechanisms to get all its command line options, but they require calling Java APIs from Ruby's API set. Rubinius provides its own API for accessing comand-line options, but I do not know if it includes VM-level flags as well as standard Ruby flags. I know there is a (({RubyVM})) namespace in the 2.0 line. If that namespace is intended to be general-purpose for VM-level features, it would be a good host for this API. Something like... ``` class << RubyVM def vm_args; end # returns array of command line args *not* passed to the target script def script; end # returns the script being executed...though this overlaps with $0 def script_args; end # returns args passed to the script...though this overlaps with ARGV, but that is perhaps warranted since ARGV can be modified (i.e. you probably want the original args) end ``` -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117874] [Ruby master Feature#6648] Provide a standard API for retrieving all command-line flags passed to Ruby
by nobu (Nobuyoshi Nakada) 14 May '24

14 May '24

Issue #6648 has been updated by nobu (Nobuyoshi Nakada). `RbConfig` is for build time informations, and does not look a right place for runtime informations. ---------------------------------------- Feature #6648: Provide a standard API for retrieving all command-line flags passed to Ruby https://bugs.ruby-lang.org/issues/6648#change-108286 * Author: headius (Charles Nutter) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- Currently there are no standard mechanisms to get the flags passed to the currently running Ruby implementation. The available mechanisms are not ideal: * Scanning globals and hoping they have not been tweaked to new settings * Using external wrappers to launch Ruby * ??? Inability to get the full set of command-line flags, including flags passed to the VM itself (and probably VM-specific) makes it impossible to launch subprocess Ruby instances with the same settings. A real world example of this is "((%bundle exec%))" when called with a command line that sets various flags, a la ((%jruby -Xsome.vm.setting --1.9 -S bundle exec%)). None of these flags can propagate to the subprocess, so odd behaviors result. The only option is to put the flags into an env var (((|JRUBY_OPTS|)) or ((|RUBYOPT|))) but this breaks the flow of calling a simple command line. JRuby provides mechanisms to get all its command line options, but they require calling Java APIs from Ruby's API set. Rubinius provides its own API for accessing comand-line options, but I do not know if it includes VM-level flags as well as standard Ruby flags. I know there is a (({RubyVM})) namespace in the 2.0 line. If that namespace is intended to be general-purpose for VM-level features, it would be a good host for this API. Something like... ``` class << RubyVM def vm_args; end # returns array of command line args *not* passed to the target script def script; end # returns the script being executed...though this overlaps with $0 def script_args; end # returns args passed to the script...though this overlaps with ARGV, but that is perhaps warranted since ARGV can be modified (i.e. you probably want the original args) end ``` -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117873] [Ruby master Feature#6648] Provide a standard API for retrieving all command-line flags passed to Ruby
by matz (Yukihiro Matsumoto) 14 May '24

14 May '24

Issue #6648 has been updated by matz (Yukihiro Matsumoto). If `RbConfig` is a convenient place for you, it is OK to add `ruby_args`. Matz. ---------------------------------------- Feature #6648: Provide a standard API for retrieving all command-line flags passed to Ruby https://bugs.ruby-lang.org/issues/6648#change-108285 * Author: headius (Charles Nutter) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- Currently there are no standard mechanisms to get the flags passed to the currently running Ruby implementation. The available mechanisms are not ideal: * Scanning globals and hoping they have not been tweaked to new settings * Using external wrappers to launch Ruby * ??? Inability to get the full set of command-line flags, including flags passed to the VM itself (and probably VM-specific) makes it impossible to launch subprocess Ruby instances with the same settings. A real world example of this is "((%bundle exec%))" when called with a command line that sets various flags, a la ((%jruby -Xsome.vm.setting --1.9 -S bundle exec%)). None of these flags can propagate to the subprocess, so odd behaviors result. The only option is to put the flags into an env var (((|JRUBY_OPTS|)) or ((|RUBYOPT|))) but this breaks the flow of calling a simple command line. JRuby provides mechanisms to get all its command line options, but they require calling Java APIs from Ruby's API set. Rubinius provides its own API for accessing comand-line options, but I do not know if it includes VM-level flags as well as standard Ruby flags. I know there is a (({RubyVM})) namespace in the 2.0 line. If that namespace is intended to be general-purpose for VM-level features, it would be a good host for this API. Something like... ``` class << RubyVM def vm_args; end # returns array of command line args *not* passed to the target script def script; end # returns the script being executed...though this overlaps with $0 def script_args; end # returns args passed to the script...though this overlaps with ARGV, but that is perhaps warranted since ARGV can be modified (i.e. you probably want the original args) end ``` -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117872] [Ruby master Feature#6648] Provide a standard API for retrieving all command-line flags passed to Ruby
by Eregon (Benoit Daloze) 14 May '24

14 May '24

Issue #6648 has been updated by Eregon (Benoit Daloze). @matz Do you agree with `RbConfig.ruby_args`, is it OK to add it? ---------------------------------------- Feature #6648: Provide a standard API for retrieving all command-line flags passed to Ruby https://bugs.ruby-lang.org/issues/6648#change-108284 * Author: headius (Charles Nutter) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- Currently there are no standard mechanisms to get the flags passed to the currently running Ruby implementation. The available mechanisms are not ideal: * Scanning globals and hoping they have not been tweaked to new settings * Using external wrappers to launch Ruby * ??? Inability to get the full set of command-line flags, including flags passed to the VM itself (and probably VM-specific) makes it impossible to launch subprocess Ruby instances with the same settings. A real world example of this is "((%bundle exec%))" when called with a command line that sets various flags, a la ((%jruby -Xsome.vm.setting --1.9 -S bundle exec%)). None of these flags can propagate to the subprocess, so odd behaviors result. The only option is to put the flags into an env var (((|JRUBY_OPTS|)) or ((|RUBYOPT|))) but this breaks the flow of calling a simple command line. JRuby provides mechanisms to get all its command line options, but they require calling Java APIs from Ruby's API set. Rubinius provides its own API for accessing comand-line options, but I do not know if it includes VM-level flags as well as standard Ruby flags. I know there is a (({RubyVM})) namespace in the 2.0 line. If that namespace is intended to be general-purpose for VM-level features, it would be a good host for this API. Something like... ``` class << RubyVM def vm_args; end # returns array of command line args *not* passed to the target script def script; end # returns the script being executed...though this overlaps with $0 def script_args; end # returns args passed to the script...though this overlaps with ARGV, but that is perhaps warranted since ARGV can be modified (i.e. you probably want the original args) end ``` -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117868] [Ruby master Feature#15554] warn/error passing a block to a method which never use a block
by Eregon (Benoit Daloze) 14 May '24

14 May '24

Issue #15554 has been updated by Eregon (Benoit Daloze). We are discussing how to make opt-in. Because relaxed mode can miss cases (e.g. for methods with a common name), and some newer codebases/some Rubyists might want to use the stricter check (e.g. it enhances readability, ignored blocks by duck typing are ignored explicitly). One idea is command-line flag: `--strict-unused-block-warning` / `--pedantic-warnings` ? Or it could be a new warning category, e.g. `:pedantic`, disabled by default (similar to `performance` category). Maybe the name should be more specific? (e.g. `:strict_unused_block`). Or not? Thoughts on this? ---------------------------------------- Feature #15554: warn/error passing a block to a method which never use a block https://bugs.ruby-lang.org/issues/15554#change-108278 * Author: ko1 (Koichi Sasada) * Status: Assigned * Assignee: matz (Yukihiro Matsumoto) ---------------------------------------- # 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` doesn`t warn warning even if this method doesn't use a block. The rule (3) can pass blocks easily and there are many methods don`t 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 (commit: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/… 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-pat… 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#… 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-pat… 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? -- https://bugs.ruby-lang.org/

1 0

[ruby-core:117838] [Ruby master Bug#20485] Simple use of Mutex and Fiber makes GC leak objects with singleton method
by skhrshin (Shintaro Sakahara) 13 May '24

13 May '24

Issue #20485 has been reported by skhrshin (Shintaro Sakahara). ---------------------------------------- Bug #20485: Simple use of Mutex and Fiber makes GC leak objects with singleton method https://bugs.ruby-lang.org/issues/20485 * Author: skhrshin (Shintaro Sakahara) * Status: Open * ruby -v: ruby 3.2.3 (2024-01-18 revision 52bb2ac0a6) [x86_64-linux] * Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN ---------------------------------------- I found a possible memory leak which occurs only when several conditions are met. The code to reproduce the problem is below: ``` class Work def add_method singleton_class.define_method(:f) {} end end Mutex.new.synchronize { Fiber.new {}.resume } work = Work.new work.add_method work = nil GC.start num_objs = ObjectSpace.each_object.select { |o| o.is_a?(Work) rescue false }.size unless num_objs.zero? raise "NG" end ``` Expected result: The script exits normally. Actual result: RuntimeError "NG" is raised. If I remove `Mutex.new.synchronize { Fiber.new {}.resume }` or `work.add_method`, GC works as expected. Is there any problem at the way to use Mutex and Fiber in this code, or is it a bug due to Ruby? I tested ruby 3.3.1 (2024-04-23 revision c56cd86388) [x86_64-linux] too and the result was same. -- https://bugs.ruby-lang.org/

2 13