Since now every functionality that changes in Bundler 4 is under a
setting, we can enable that setting to test the new functionality,
without having to run our full CI twice.
This can actually be seen as increasing coverage, because Bundler 4
functionality will now be tested on Windows, MacOS, or any other
environment where previously "Bundler 4 mode" was not running.
1cb3e009fc
If you have
```
gem "rails", git: "https://github.com/rails/rails"
```
and then explicitly pin to an older ref, like
```
gem "rails", git: "https://github.com/rails/rails", ref: "99bacb5aa8"
```
Then `bundle install` fails, because locked sources fail to be updated
to use the new source.
This commit fixes the problem by making sure get their source properly
replaced.
5de8c2e0cf
* Increase timeout to fix flaky tests?
* Fix bundler test - wording changed.
expect "fatal: Remote branch deadbeef not found in upstream origin" to
include "Revision deadbeef does not exist in the repository"
This has the following benefits:
* Avoid duplicated work in some specs that first build a repo, and then
overwrite it with a completely different set of gems.
* Reduce RSpec nesting and improve readability.
* The change also made surfaces several specs that were incorrect since
they were unintentionally not testing the right thing.
ed430883e0
`version` is actually an attribute of the dependency, not of the git
source. Sometimes it's passed to the git source to be able to fake a
gemspec in case there's no gemspec in the source, but it should not be
used for source comparison.
d936fbd78e
Specs that use extension gems were failing in the new job but I noticed
that they were using very non standard `extconf.rb` files.
The hack being removed here was added just to make specs pass when run
in ruby-core but it seems the underlying issue has been fixed now, and
it's causing issues with Ruby 3.3 and Windows, so necessary so I'm
removing it and moving on.
5b78275f0e
Previously, when sorting and comparing git Gemfile vs lockfile sources during
`bundler/setup` to figure out whether we need to re-resolve or not, we
would try to find the default branch if nothing more specific was
specified in the Gemfile.
If the git cache has been deleted thought, that would fail.
The error would still be swallowed (and the branch would simply not be
displayed), but trying to clone would still generate the side effect of
creating the parent folder for the clone.
That could affect non-writable systems that don't expect `bundler/setup`
to write to the filesystem at all.
To fix this, override `Bundler::Source::Git#identifier` to use
exclusively static information, so it does not even try to clone the
repo nor generate any side effects.
582eb2ef39
When testing under the ruby/ruby setup, mkmf.rb needs to the `$extout`
global variable set properly.
This is because, in this particular case, the `ruby.h` header needed to
compile extensions is constructed from
`$(extout)/include($arch)/ruby/config.h` but `$extout` is not set by
default.
I tried to fix this in mkmf.rb itself but I couldn't figure it. But
setting it externally to workaround the issue fixes the specs, so I'll
start with that. Also setting it externally causes issues when running
specs upstream against Ruby 2.3 (I guess because of some difference with
Ruby 2.3 mkmf.rb implementation). So I'm avoiding doing it on Ruby 2.3 to
woraround that.
d782984585
This doesn't affect the outcome of the test, but it makes the `git
commit` command inside `update_git` not fail because of not having
anything to commit.
ad0160ed97
