Feature #12719: `Struct#merge` for partial updates

• Status: Feedback
• Assignee: -
• Target version: -

[halogenandtoast (Matthew Mongeau)]

Other languages have operators for performing partial updates on maps. I feel like Struct could be more useful if it provided an easy way of performing partial (or full) updates.

After the change you can do the following:

Point = Struct.new(:x, :y)

p = Point.new(1, 2)
p2 = p.merge(y: 4)
p3 = p2.merge(x: 10)

p.inspect     # => #<struct Point x=1, y=2>
p2.inspect   # => #<struct Point x=1, y=4>
p3.inspect   # => #<struct Point x=10, y=4>

p.merge!("x" => 9)

p.inspect    # => #<struct Point x=9, y=2>

[halogenandtoast (Matthew Mongeau)]

• Tracker changed from Bug to Feature

[halogenandtoast (Matthew Mongeau)]

If this is well received I think a similar syntax could be used for hashes in place of merge.

[halogenandtoast (Matthew Mongeau)]

• File struct_merge.patch added

As an alternative since the | syntax might get shot down. Here's a patch adding a merge function instead:

Point = Struct.new(:x, :y)

p = Point.new(1, 2)
p2 = p.merge(y: 4)
p3 = p2.merge(x: 10)

puts p.inspect # => #<struct Point x=1, y=2>
puts p2.inspect # => #<struct Point x=1, y=4>
puts p3.inspect # => #<struct Point x=10, y=4>

[nobu (Nobuyoshi Nakada)]

In your example, the value in the LHS is ignored when the same key is present in the RHS hash.
It doesn't feel nice as | operator.

merge sounds nice in that sense, but your patch would segfault at p.merge(0).

[halogenandtoast (Matthew Mongeau)]

Thanks, nice catch. I'll update this tomorrow to not segfault.

[halogenandtoast (Matthew Mongeau)]

• File struct_merge_no_segfault.patch added

Updated so it won't segfault

[halogenandtoast (Matthew Mongeau)]

• File merge_bang.patch added

Since merge closely resembles the similar hash function, I think it makes sense to also add merge! as a function. I'm not a fan of the mutating methods, but I would find it surprising if this interface was different. Here's an updated patch. I also fixed some of the documentation I wrote.

[halogenandtoast (Matthew Mongeau)]

• File deleted (struct_update.patch)

[halogenandtoast (Matthew Mongeau)]

• File deleted (struct_merge.patch)

[halogenandtoast (Matthew Mongeau)]

• File deleted (struct_merge_no_segfault.patch)

[halogenandtoast (Matthew Mongeau)]

• File struct_merge.patch struct_merge.patch added
• Subject changed from `Struct#|` for partial updates to `Struct#merge` for partial updates
• Description updated (diff)

Update ChangeLog

[halogenandtoast (Matthew Mongeau)]

• File deleted (merge_bang.patch)

[matz (Yukihiro Matsumoto)]

• Status changed from Open to Feedback

I want to see a real-world use-case for the feature.
In addition, I don't think the name merge is the best for the functionality.

Matz.

[Eregon (Benoit Daloze)]

Scala has "copy" for this purpose: some_case_class_object.copy(field: new_value)

[marcotc (Marco Costa)]

To add a real-world use-case: in the ddtrace gem we have a configuration Struct called AgentSettings that holds the global defaults for our configuration:

AgentSettings = Struct.new(
  :ssl,
  :hostname,
  :port,
  :timeout_seconds,
  :deprecated_for_removal_transport_configuration_proc,
  :deprecated_for_removal_transport_configuration_options
) do
  def initialize(
    ssl:,
    hostname:,
    port:,
    timeout_seconds:,
    deprecated_for_removal_transport_configuration_proc:,
    deprecated_for_removal_transport_configuration_options:,
  )
    super(ssl, hostname, port, timeout_seconds, deprecated_for_removal_transport_configuration_proc, \
              deprecated_for_removal_transport_configuration_options)
    freeze
  end
end

But in a few places we want to override some of the default configuration values:

transport.adapter(
  default_adapter,
  agent_settings.hostname,
  agent_settings.port,
  # We explicitly use profiling_upload_timeout_seconds instead of agent_settings.timeout because profile
  # uploads are bigger and thus we employ a separate configuration.
  timeout: profiling_upload_timeout_seconds,
  ssl: agent_settings.ssl
)

Today we resort to unpacking the Struct and effectively transferring the values to a hash (in the form of keyword arguments). We don't actually want our Struct to become a hash (or keyword arguments) when we pass it to #adapter in the example above, we would like a AgentSettings struct to be passed but with timeout overridden.