Skip to content

Run specs in parallel, speed up CI, fix concurrency bugs in specs and code#1167

Draft
apiology wants to merge 1562 commits intocastwide:v0.59from
apiology:parallel_rspec
Draft

Run specs in parallel, speed up CI, fix concurrency bugs in specs and code#1167
apiology wants to merge 1562 commits intocastwide:v0.59from
apiology:parallel_rspec

Conversation

@apiology
Copy link
Contributor

@apiology apiology commented Jan 29, 2026

This allows rspec to use more than one CPU. Really useful on modern dev machines, and also useful in CI, where the default runner size has two CPUs.

See https://github.com/apiology/solargraph/pull/27/changes for a more useful diff and commentary

To do:

  • Track down slow tests and see where time is going
  • Track down rspec + rails specs slowness
  • Add developer doc on things to expect, how to write specs
  • Fix logging caching gems
  • Figure out where doc_map vs workspace fix went

Reduces the actual testing time in CI to about a minute using our small workers:

image

There are also a few fixes here:

  • For some reason, the debug level logging I tried to set in rspec wasn't persisting - this seems to do a better job and found a couple of bugs in the process that only reproduce at debug level logging.
  • Some tests didn't run individually, but did when run in sequence in the specs. We may see new failures pop up in the future; the trade-off for faster responsiveness is probably going to be less deterministic behavior in the face of buggy tests.

Includes for mergability:

Includes this build fix:

Includes the fix for this bug, which parallel running seems to trigger (more often?):

Includes to show combined effect:

Includes transitively:

  • v0.59 branch

apiology and others added 30 commits January 13, 2026 18:22
Some hand-changed fixes for RuboCop issues
* Use each_line instead of text.lines

Avoid allocating additional strings, instead use sliced substrings

* gitignore vendor/cache

* Remove redundant end_with?

Co-authored-by: Jean Boussier <jean.boussier@gmail.com>

* Remove benchamrks

castwide#1054 (comment)

* String#index(offset:) FTW 🚀

castwide#1054 (comment)

Co-authored-by: Jean Boussier <jean.boussier@gmail.com>

* fix rubocop

---------

Co-authored-by: Jean Boussier <jean.boussier@gmail.com>
* Raise InvalidOffsetError for offsets > text

* Linting

* Fix fencepost error

* Additional fencepost test

* Document exception
On my machine, I got a 33% speed-up with these tweaks:

before: Finished in 1 minute 15.88 seconds
after: Finished in 49 seconds
* Fix merge

* Adjust annotations

* Fix typecheck errors

* Fix merge

* Fix merge

* Fix some @sg-ignores

* Fix merge

* Merge branch 'intersection_types' into flow_sensitive_typing_2_0

* Fix RuboCop issue

* Fix type issues

* Fix type issues

* Fix rspec.yml

* Fix rspec.yml

* Add @sg-ignores

* Remove @sg-ignores

* Merge branch 'or_support_in_flow_sensitive_typing' into union_type_enforcement

* Fix spec

* Fix annotations

* Fix RuboCop issues

* Bump RBS versions in rspec test

* Fix version

* Fix version matrix

* Fix version matrix

* Fix version matrix

* Fix version matrix

* Fix version matrix

* Exclude another

* Exclude another

* Add version, fix doc

* init -> config

* Fix rbs-version for Ruby 4.0 in CI workflow

* Clean up ruby-version entries in rspec.yml

Removed deprecated ruby-version entries for RBS.

* Fix RBS version for Ruby 4.0 in workflow

* Fix merge

* Improve signature combination

Use our generated RBS signature from parameters as a key to combine
method signatures from RBS/YARD pins.

This is closer to what RBS does than the current technique of using
the arity alone, and fixes a key degenerate case in Integer#+ revealed
by updated definitions used by recently released RBS gems

* Update annotations

* Drop annotation

* Fix RuboCop issue

* Fix merge

* Don't use solargraph-rspec branch

* Fix merge

* Debug

* Add another use of stdlib dependencies in RBS

* Mock additional call

* Fix annotations

* Update types in rspec undercover

* Debug

* Debug

* Drop incorrect rbs collection use in spec

* Update rubocop todo

* Revert change

* Fix RuboCop issue

* Fix annotations

* Fix annotations

* RuboCop fix

* Use "type arity" to guide signature combination

* Update rubocop todo

* Include return type arity in comparison

* Add dodgy return type

* Fix RuboCop issue

* Add Ruby 4.0 jobs

* Exclude another combo

* Exclude another combo

* Update rules to use report?

* Fix merge

* Fix merge

* Drop dead code

* Bump version to 0.59.0.dev.1

* Rename rule

* Update RuboCop todo file

* Update RuboCop todo file

* Ratchet rubocop TODO file

* Move to skip:

* Mark spec as pending

* Revert spec change

* Drop old workaround

* Fix merge

* Fix typechecking issues

* Revert doc

* Fix spelling

* Fix merge issue

* Exclude the current gemspec from pins brought in from gem

* Check pathname instead

* Add sg-ignore

* Avoid rbs pollution

We were using the sig/shims directory for some internally helpful
shims; unfortunately that exported them during gem installs, causing
castwide#1144

* Test with RBS 4.0.0.dev.5

* Open up in gemspec

* Fix missing spot

* Typecheck using RBS prereleases

* Move point of ignoring cached gems for gem projects

* Fix issues resolving cgi escape functions

* Be more careful marking things as stdlib

* Reclassify rbs gem

* Fix merge

* Add sg-ignore

* Fix merge

* Remove outdated workaround

* Fix @sg-ignore name

* Restore workaround

* Restore workaround

* Merge branch 'flow_sensitive_typing_2_0' into 2025-01-06

* Fix method signature

* Fix annotations

* Add regression test and fix for issue found during future merge

* Add regression test and fix for issue found during future merge

* Fix merge

* Fix merge

* Fix merge

* Fix merge

* Use correct field for self type resolution

Add a regression test and fix for self type resolution issue found on
a future branch

* Fix 'solargraph pin --references ClassName' private method call

* Add error handling

* Fix another location with another test case

* Drop now-unneeded @sg-ignore

* Don't log caching for each dependent library

This causes duplicate logging on standard libraries, many of which are
esoteric (e.g., "cgi-escaping").  The current method as of the
2025-01-06 branch would result in each stdlib library being cached
individually.

* Drop logging entirely

* Fix some types based on future branch feedback

* Provide Gem::Specification to outside interface

* Provide Gem::Specification to outside interface

* Use #to_spec

* Provide Gem::Specification to outside interface

* Fix typechecking error

* Use consistent bundler versions

* Fix type issue

* Fix annotations based on future branch feedback

* Add some @todos

* Fix annotations

* Fix annotations

* Fix annotation

* Add diff::lcs shim

* Improve spec expectations

* Add @sg-ignore

* Fix rspec checks to run on all types of PRs

* Fix merge

* Fix merge

* Fix merge
Let's work more directly with type objects and reduce duplication
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants