Accept terrapin 1.x

[tricknotes]

The gemspec pinned terrapin ~> 0.6.0 — a version from February 2018. terrapin has been at 1.x since November 2023.

-  spec.add_dependency "terrapin", "~> 0.6.0"
+  spec.add_dependency "terrapin", ">= 0.6.0", "< 2.0"

Why 1.0 is safe here

Its only breaking change was the removal of one class:

Terrapin::CommandLine::PosixRunner was removed. You can replace any usage of this with Terrapin::CommandLine::ProcessRunner, which uses Ruby's builtin Process.spawn.

Nothing here uses a runner. Every use of terrapin in this gem builds a command line and stops there — Command calls Terrapin::CommandLine#command and hands the resulting string to Runner (Open3) or Kernel.spawn:

lib/ember_cli/command.rb:13   line.command
lib/ember_cli/command.rb:35   line.command(host: …, port: …)
lib/ember_cli/command.rb:78   line.command(mode: …, output_path: …)
lib/ember_cli/command.rb:102  line.command(environment: …, output_path: …, watcher: …)

Against terrapin 1.1.1 those strings come out as they did under 0.6.0, shell quoting of interpolated values included:

/path/to/vite build --mode 'production' --outDir '/tmp/out dir' --emptyOutDir
/path/to/ember test --environment test

The pin reached past this gem

terrapin 0.6.0 requires climate_control >= 0.0.3, < 1.0. Since this gem pinned terrapin to 0.6, every application depending on ember-cli-rails was held below climate_control 1.0, with no way out short of dropping the gem. terrapin 1.x requires only climate_control >= 0; installing 1.1.1 here pulled climate_control from 0.2.0 to 1.2.0.

Tests

With terrapin 1.1.1 and climate_control 1.2.0 resolved locally:

• bin/rspec spec/lib — 174 examples, 2 failures, both (app_spec.rb:192, app_spec.rb:200) failing identically on main; they need node_modules installed to find the ember binary
• bin/rspec spec/lib/ember_cli/command_spec.rb — 15 examples, 0 failures

One thing to note

Gemfile.lock is not committed here, so once this lands CI resolves terrapin to the newest allowed version and 0.6 stops being exercised. The constraint still permits it for applications that have it locked. If keeping 0.6 under test matters, a matrix axis pinning it would be the way, and I can add one.