Update specs for 3.4 behavior changes

[headius]

ruby/spec is currently not being run on Ruby 3.4, and as a result there are a number of failures. Most of these are related to the new Hash#inspect formatting (preferring a: 1 over :a => 1 with some spacing changes) but there are others.

We will need to update the specs and add version guards for cosmetic changes like this, and audit the remaining specs for real behavior changes that should be guarded or added.

[eregon]

Ruby 3.4 is not yet released, so it doesn't make sense to test here against it as it keeps changing.
ruby/spec is run in ruby/ruby's CI.
So basically this will be solved the next time specs are synchronized by @andrykonchin.
https://github.com/ruby/spec?tab=readme-ov-file#synchronization-with-ruby-implementations

[eregon]

For example see ruby/ruby@c94815b which will be mirrored here at the next synchronization.

[headius]

this will be solved the next time specs are synchronized

Sounds good.

[eregon]

If you mean creating an item list like #1016 for 3.3 or 3.4, feel free to create it.
It's just the NEWS file with some extra text at the beginning and some reformatting so the checkboxes are at the top-level.

[headius]

We are interested in syncing the specs sooner than whatever schedule @andrykonchin does it. I'm reading through the instructions for upstreaming and downstreaming now.

[headius]

I attempted to sync from MRI and got an error about the last merge being too long ago?

To /Users/headius/work/rubyspec
 * [new branch]            specs-2024-10-31 -> mri
git checkout master
Already on 'master'
Your branch is up to date with 'origin/master'.
git pull
Already up to date.
git checkout mri
Switched to branch 'mri'
Last merge is 747957569427f35043aa74a4df610339464a9dad
sync-rubyspec.rb:156:in `block in rebase_commits': 121 days since last merge, probably wrong commit (RuntimeError)
	from sync-rubyspec.rb:126:in `chdir'
	from sync-rubyspec.rb:126:in `rebase_commits'
	from sync-rubyspec.rb:233:in `block in main'
	from sync-rubyspec.rb:229:in `each_pair'
	from sync-rubyspec.rb:229:in `main'
	from sync-rubyspec.rb:254:in `<main>'

This is with mspec and rubyspec as sibling directories both on master and following the instructions at https://github.com/ruby/spec/wiki/Merging-specs-from-JRuby-and-other-sources

[eregon]

There is an env var to skip that check, see the source of that script.

[headius]

Yes, I saw that, but has it really been that long since the specs were synced? If so that's fine, I just thought I did something wrong.

[eregon]

Yes, the last sync was on July 1: ruby/ruby#11080 jruby/jruby#8311

[headius]

I made an attempt here but a huge amount of pack and unpack specs are failing. They appear to have no version guards.

[headius]

Ahh I see, there are mspec changes that need to be synced as well. I'll try to pick this up tomorrow.

[eregon]

Yes, MSpec should be sync'd first as documented in https://github.com/ruby/spec/wiki/Merging-specs-from-JRuby-and-other-sources#upstream-mspec-changes-in-mri-jruby-and-truffleruby
Please do all the steps as documented there (in order), otherwise it might be hard to fix it or understand what's going on with some later issue.

[headius]

The first thing the document does is list the order in which things should be synced:

Synchronizing specs between the ruby/spec repository and Ruby/JRuby/TruffleRuby should be performed in the following order:

• pushing changes made in Ruby implementations to ruby/spec (upstreaming)
• pulling ruby/spec changes back from ruby/spec to Ruby implementations (downstreaming)

So that is where I started. Perhaps the document should be updated to make it clear what the actual sequence of steps is supposed to be.

[eregon]

That's the general overview of the approach, the expectation is steps are done in the order they are written.
Feel free to update the wiki if it's not clear.

[headius]

It appears the issues were not entirely my fault. CRuby has made backward-incompatible changes to mspec that prevent current specs from passing: ruby/mspec#70

Specifically, they changed the way that some platform details are guarded, for example the platform bit width for native numeric types. Without mspec and rubyspec updating at the same time it's not possible to pass CI on either side.

Feel free to update the wiki if it's not clear.

Done. I will probably have additional edits.