Upcoming Puppet 4 / Ruby 2.1.5 Registry compatibility changes

[Iristyle]

As part of making Puppet 4 compatible with Ruby 2.1.5, we have discovered a couple of bugs inside Ruby that will affect your module. The two primary issues are:

• When enumerating keys with #each_key / #keys or values with #each_value / #values, Ruby will take a UTF-16LE string and roundtrip it through the local codepage unnecessarily, which may cause encoding exceptions. Our prime example of this is when a Unicode en-dash U+2013 appears in a registry value, and the local codepage is IBM437, which has no equivalent character. This is just one of many potential examples that may trigger this behavior. NOTE: Normal registry reads of a value at a particular key are not problematic - the bad behavior is only triggered during enumeration. We have not yet filed this bug with Ruby.
• When using the #delete_key and #delete_value methods, Ruby is passing wide character UTF-16LE strings to ANSI APIs. We have filed this bug with Ruby at https://bugs.ruby-lang.org/issues/10820

There are additional technical details that can be found in PUP-3837, the ticket we're using to track changes that we've made to Puppet. Our resolution for these issues was merged to Puppet at https://github.com/puppetlabs/puppet/commit/c610cd01eeef3fafa7aa2761a3435dd6c1b0d8d4

We don't have Puppet 4 builds released yet, but keep an eye on https://groups.google.com/forum/#!forum/puppet-announce

I have identified issues with the code in this module that will require some updating to ensure it maintains compatibility with both existing Puppet releases and the upcoming releases. I'm getting in touch with you now, so that you're not caught off guard, and so that your module is ready to go by the time 4.0 ships.

Issues:

• each is called at https://github.com/liamjbennett/puppet-win_facts/blob/master/lib/facter/operatingsystemversion.rb#L18, https://github.com/liamjbennett/puppet-win_facts/blob/master/lib/facter/windows_productkey.rb#L55 and https://github.com/liamjbennett/puppet-win_facts/blob/master/lib/facter/windows_sid.rb#L22, which will trigger the errant Ruby behavior. Instead, use an internal FFI definition for the RegEnumKeyW and RegEnumValueW API calls like we've done inside Puppet at https://github.com/puppetlabs/puppet/blob/c610cd01eeef3fafa7aa2761a3435dd6c1b0d8d4/lib/puppet/util/windows/registry.rb#L89-L149
• You might want to also verify the module by hand using a Ruby 2.1.5 runtime. To do this requires modifying registry.rb, present at c:\tools\ruby215\lib\ruby\2.1.0\win32\registry.rb in my install. Some simple warnings may be emitted in several locations where problematic behavior can occur. Then you can exercise your module / run specs for a full validation.
  • https://github.com/ruby/ruby/blob/v2_1_5/ext/win32/lib/win32/registry.rb#L815 should have puts "BUG: RegDeleteValue called with #{name} against #{@keyname}"
  • https://github.com/ruby/ruby/blob/v2_1_5/ext/win32/lib/win32/registry.rb#L836 and https://github.com/ruby/ruby/blob/v2_1_5/ext/win32/lib/win32/registry.rb#L841 should have puts "BUG: RegDeleteKey called with #{name} against #{@keyname}" added
  • https://github.com/ruby/ruby/blob/v2_1_5/ext/win32/lib/win32/registry.rb#L895 should have puts "BUG: export_string triggered with #{str} to #{enc.to_s}"
  • Optionally https://github.com/ruby/ruby/blob/v2_1_5/ext/win32/lib/win32/registry.rb#L363 can have puts "BUG: expand_environ triggered with #{str}" - while this behavior does trigger a string re-encode to the current codepage, we view it as less problematic as it requires REG_MULTI_SZ and environment variable expansion - where the values will typically be in the current codepage on a default system.

We will be making an announcement to the puppet-dev list shortly mentioning some of these upcoming changes. https://groups.google.com/forum/#!forum/puppet-dev

Let me know if you have any questions.

Thanks!

[Iristyle]

Also note that because this module is a dependency for liamjbennett/windows_power, it is also affected by association.