Experimental removal of Rubygems workarounds (main branch) #2380
Conversation
pirj
force-pushed
the
reckless-rubygems-workaround-removal
branch
from
0d5fee8
to
35a6b21
Compare
pirj
changed the title
|
Ouhao. Super happy if it is working. |
pirj
force-pushed
the
reckless-rubygems-workaround-removal
branch
2 times, most recently
from
d0937af
to
e09de36
Compare
|
Ran the build several times, it's green apart from a
|
|
Have we seen the mountable_engine? failure on main yet? It must be related to #2372 |
Gemfile
Outdated
| if MAJOR >= 6 | ||
| gem 'selenium-webdriver', '~> 3.5', require: false | ||
| # Rails 6.0 specify '~> 1.4' directly in the adapter. |
There was a problem hiding this comment.
Do you mean > 1.4? Either way we should let Rails decide what version to use?
There was a problem hiding this comment.
The comment below makes no sense. When sqlite3 1.5.0 is out, it still matches the Rails' requirement.
But if it happens that sqlite3 decides to jump straight to 2.0, our builds will be broken if we omit this constraint.
I'll fix the comment.
There was a problem hiding this comment.
Sorry it was late when I wrote this, what I mean is if the Rails constraint is ~> 1.4 then this has no effect and can be removed. If the rails constraint is > 1.4 then this has an effect but again I don't feel we should constrain this, we need to match Rails here.
There was a problem hiding this comment.
Point is sqlite3 is an optional Rails dependency, they only call gem when the adapter is set to sqlite in config/database.yml. But still they have a version constraint defined there, which Bundler has no awareness of, since it's outside of the Gemfile.
If we remove it, sqlite3 won't be installed and the build would break.
If we remove the version constraint, the build will break for Rails 6.0 when (eventually, far in the future) sqlite3 2.0 is out.
There was a problem hiding this comment.
Right ok, thats not clear from the comment at all.
pirj
force-pushed
the
reckless-rubygems-workaround-removal
branch
from
efb680e
to
5dae54f
Compare
Gemfile
Outdated
| if MAJOR >= 6 | ||
| gem 'selenium-webdriver', '~> 3.5', require: false | ||
| # Rails 6.0 specify '~> 1.4' directly in the adapter. |
There was a problem hiding this comment.
Sorry it was late when I wrote this, what I mean is if the Rails constraint is ~> 1.4 then this has no effect and can be removed. If the rails constraint is > 1.4 then this has an effect but again I don't feel we should constrain this, we need to match Rails here.
pirj
force-pushed
the
reckless-rubygems-workaround-removal
branch
from
1dc7b46
to
cde18a4
Compare
|
All green, except for |
With the recent Rubygems server fixes and database backfills, we don't need most of the manual workarounds to resolve dependency Ruby version support.
13 is out, no reason to stick to 12
pirj
force-pushed
the
reckless-rubygems-workaround-removal
branch
from
cde18a4
to
09e465c
Compare
| if MAJOR >= 6 | ||
| gem 'selenium-webdriver', '~> 3.5', require: false | ||
| # sqlite3 is an optional, unspecified, dependency and Rails 6.0 only supports `~> 1.4` | ||
| gem 'sqlite3', '~> 1.4', platforms: [:ruby] |
There was a problem hiding this comment.
example_app_generator/generate_app.rb ends up with
gem 'sqlite3'in example_app's Gemfile. Why there's no version mismatch, and how the proper version is resolved is beyond my understanding.
I used a different approach for 3-9-maintenance branch, moved this selection to Gemfile-rails-dependencies, and added a statement to strip example_app's Gemfile from gem 'sqlite3'. https://github.com/rspec/rspec-rails/pull/2386/files#diff-8c3955be04b733dbd31e54e50844fb84
PS and it didn't work for Rails 4 https://travis-ci.org/github/rspec/rspec-rails/jobs/734530137
Experimental removal of Rubygems workarounds (main branch)
Due to unstable version resolution and lack of Ruby version constraints on older gemspecs in the Rubygems database, we had to manually declare versions to be used.
This has seemingly been fixed.
Related:
rubygems/rubygems#3463
rubygems/rubygems.org#2474
rubygems/rubygems#2662
Let's see if it works now, and if the issue with IPv6 fallback is still the case.