[Source::Path] Sort gemspecs by relative depth

[segiddins]

Attempts to close #3286, https://trello.com/c/DQx0OKJl/82-search-for-gemspecs-near-the-root-of-a-git-repo-first

\c @TimMoore

[indirect]

hah, clever. I was going to have two globs, the first only matching the top level. I think this is a better general solution

[segiddins]

Let me know if you want specs for this change

-Samuel E. Giddins

On Feb 23, 2015, at 6:25 PM, André Arko [email protected] wrote:

hah, clever. I was going to have two globs, the first only matching the top level. I think this is a better general solution

—
Reply to this email directly or view it on GitHub.

[indirect]

I think the existing specs probably cover this sufficiently? Of course, I may regret saying that in a few weeks or months, so if you could add a spec that would be great. 👍

[segiddins]

Hm, I can't use the premailer example to make a realworld spec, since only their development dependencies differ.

[TimMoore]

👍 thanks @segiddins!

A spec would be good, if only to make sure nobody breaks it again in the future. It might work to use the build_lib helper to build a simple gem with an invalid gemspec (one with just raise "FAIL" in it for example) in a subdirectory that sorts before the real one alphabetically. Then the spec can verify that the path gem still works in a Gemfile.

[segiddins]

@TimMoore would be happy to -- any pointers on how to do that with build_lib? It always just seems to be used like build_lib 'foo'

[TimMoore]

@segiddins it can take a block, and yields a builder object that can be used to write files into the path. See https://github.com/bundler/bundler/blob/master/spec/install/path_spec.rb#L56 for example

      build_lib "rack", "1.0.0", :to_system => true do |s|
        s.write "lib/rack.rb", "raise 'FAIL'"
      end

[segiddins]

@TimMoore updated with a spec + comments :)

Turned out the issue was Index#<< overriding the earlier specs -- in other words, the glob pattern was not the actual problem.

[indirect]

Yay specs! :)

On Thu, Feb 26, 2015 at 4:13 PM, Samuel E. Giddins
[email protected] wrote:

@TimMoore updated with a spec + comments :)

Turned out the issue was Index#<< overriding the earlier specs -- in other words, the glob pattern was not the actual problem.

Reply to this email directly or view it on GitHub:
#3428 (comment)

[TimMoore]

LGTM 👍