Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Prevent deep_merge from mutating nested hashes #467

Merged
merged 1 commit into from Nov 18, 2019

Conversation

michaelherold
Copy link
Member

The DeepMerge extension has two methods of mutating hashes: a
destructive one and a non-destructive one. The #deep_merge version
should not mutate the original hash or any hash nested within it. The
#deep_merge! version is free to mutate the receiver.

Without deeply duplicating the values contained within the hash, the
invariant of immutability cannot be held for the original hash. In order
to preserve that invariant, we need to introduce a method of deeply
duplicating the hash.

The trick here is that we cannot rely on a simple call to Object#dup.
Some classes within the Ruby standard library are not duplicable in
particular versions of Ruby. Newer versions of Ruby allow these classes
to be "duplicated" in a way that returns the original value. These
classes represent value objects, so it is safe to return the original
value ... unless the classes are monkey-patched, but that isn't
something we can protect against.

This implementation does a best-effort to deeply duplicate an entire
hash by relying on these value object classes being able to return
themselves without violating immutability.

Copy link
Member

@dblock dblock left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also needs a rebase.

else
value.dup
end
end
Copy link
Member

@dblock dblock Oct 2, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this a pattern from other libraries? I feel like taking extra overhead to call dup on even those simple types is OK because you can't make assumptions about any underlying implementation. It's also unclear whether there's performance gain doing a case vs. a dup, so maybe just not worth it?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The issue is that on Rubies < 2.4(? I think ... maybe 2.5), these classes raise a TypeError when you call #dup on them. Since they are all written in C, they need to have an allocate method for #dup to work and they do not.

I can roll out the safe dup change and run it on CI if you'd like to see. Otherwise, you can pull it down and try running the test suite after changing the Hashie::Utils.safe_dup(value) line to value.dup.

I originally went the route of a refinement that added ActiveSupport-like feature detection for `#dup, but decided that was a lot of overhead to do something that is just as fast and much less code.

What do you think?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, got it. I think what you did makes sense then, I didn't know this.

Copy link
Member Author

@michaelherold michaelherold left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll the get squash for the changelog once I get your thoughts on the item below.

else
value.dup
end
end
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The issue is that on Rubies < 2.4(? I think ... maybe 2.5), these classes raise a TypeError when you call #dup on them. Since they are all written in C, they need to have an allocate method for #dup to work and they do not.

I can roll out the safe dup change and run it on CI if you'd like to see. Otherwise, you can pull it down and try running the test suite after changing the Hashie::Utils.safe_dup(value) line to value.dup.

I originally went the route of a refinement that added ActiveSupport-like feature detection for `#dup, but decided that was a lot of overhead to do something that is just as fast and much less code.

What do you think?

@@ -3,7 +3,7 @@ module Extensions
module DeepMerge
# Returns a new hash with +self+ and +other_hash+ merged recursively.
def deep_merge(other_hash, &block)
copy = dup
copy = _deep_dup(self)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was wondering why we create a dup all the time? What is the reason behind this?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The reason for this is due to the semantics between #merge and #merge!. The plain method does not mutate the receiver, the bang-method does.

@michaelherold michaelherold self-assigned this Oct 4, 2019
The `DeepMerge` extension has two methods of mutating hashes: a
destructive one and a non-destructive one. The `#deep_merge` version
should not mutate the original hash or any hash nested within it. The
`#deep_merge!` version is free to mutate the receiver.

Without deeply duplicating the values contained within the hash, the
invariant of immutability cannot be held for the original hash. In order
to preserve that invariant, we need to introduce a method of deeply
duplicating the hash.

The trick here is that we cannot rely on a simple call to `Object#dup`.
Some classes within the Ruby standard library are not duplicable in
particular versions of Ruby. Newer versions of Ruby allow these classes
to be "duplicated" in a way that returns the original value. These
classes represent value objects, so it is safe to return the original
value ... unless the classes are monkey-patched, but that isn't
something we can protect against.

This implementation does a best-effort to deeply duplicate an entire
hash by relying on these value object classes being able to return
themselves without violating immutability.
@dblock dblock merged commit 4f014e7 into hashie:master Nov 18, 2019
bmwiedemann pushed a commit to bmwiedemann/openSUSE that referenced this pull request Jan 22, 2021
https://build.opensuse.org/request/show/865196
by user coolo + dimstar_suse
- updated to version 4.1.0
 see installed CHANGELOG.md
  ## [4.1.0] - 2020-02-01

  [4.1.0]: hashie/hashie@v4.0.0...v4.1.0

  ### Added

  * [#499](hashie/hashie#499): Add `Hashie::Extensions::Mash::PermissiveRespondTo` to make specific subclasses of Mash fully respond to messages for use with `SimpleDelegator` - [@michaelherold](https://github.com/michaelherold).

  ### Fixed

  * [#467](hashie/hashie#467): Fixed `DeepMerge#deep_merge` mutating nested values within the receiver - [@michaelherold](https://github.com/michaelherold).
  * [#505](hashie/hashie#505): Ensure that `Hashie::Array`s are not deconverted within `Hashie::Mash`es to make `Mash#dig` work properly - [@Michaelhero
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

3 participants