Update ruby bundle to commit 77dc5d13cc

This commit is contained in:
Kevin Sawicki 2012-10-10 13:48:46 -07:00
parent 670da2c9c3
commit f694d1d2c8
2 changed files with 7 additions and 7 deletions

View File

@ -77,7 +77,7 @@ task "clone-default-bundles" => "create-dot-atom" do
"https://github.com/textmate/html.tmbundle.git" => "HEAD", "https://github.com/textmate/html.tmbundle.git" => "HEAD",
"https://github.com/textmate/javascript.tmbundle.git" => "HEAD", "https://github.com/textmate/javascript.tmbundle.git" => "HEAD",
"https://github.com/textmate/ruby-on-rails.tmbundle.git" => "HEAD", "https://github.com/textmate/ruby-on-rails.tmbundle.git" => "HEAD",
"https://github.com/textmate/ruby.tmbundle.git" => "f855d76db13d74a8bf7489f4a7722654685756ea" , "https://github.com/textmate/ruby.tmbundle.git" => "77dc5d13ccb4feabca82dd8d9ed7889ec1061455" ,
"https://github.com/textmate/text.tmbundle.git" => "HEAD", "https://github.com/textmate/text.tmbundle.git" => "HEAD",
"https://github.com/jashkenas/coffee-script-tmbundle.git" => "HEAD", "https://github.com/jashkenas/coffee-script-tmbundle.git" => "HEAD",
"https://github.com/cburyta/puppet-textmate.tmbundle.git" => "HEAD", "https://github.com/cburyta/puppet-textmate.tmbundle.git" => "HEAD",

View File

@ -12,7 +12,7 @@ describe "TextMateGrammar", ->
describe ".getLineTokens(line, currentRule)", -> describe ".getLineTokens(line, currentRule)", ->
describe "when the entire line matches a single pattern with no capture groups", -> describe "when the entire line matches a single pattern with no capture groups", ->
it "returns a single token with the correct scope", -> it "returns a single token with the correct scope", ->
{tokens} = grammar.getLineTokens("return") {tokens} = grammar.getLineTokens("return")
expect(tokens.length).toBe 1 expect(tokens.length).toBe 1
@ -179,11 +179,11 @@ describe "TextMateGrammar", ->
{tokens} = grammar.getLineTokens('%Q+matz had some #{%Q-crazy ideas-} for ruby syntax+ # damn.') {tokens} = grammar.getLineTokens('%Q+matz had some #{%Q-crazy ideas-} for ruby syntax+ # damn.')
expect(tokens[0]).toEqual value: '%Q+', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.begin.ruby"] expect(tokens[0]).toEqual value: '%Q+', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.begin.ruby"]
expect(tokens[1]).toEqual value: 'matz had some ', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby"] expect(tokens[1]).toEqual value: 'matz had some ', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby"]
expect(tokens[2]).toEqual value: '#{', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","meta.embedded.line.ruby","punctuation.section.embedded.begin.ruby"] expect(tokens[2]).toEqual value: '#{', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","source.ruby.embedded.source","punctuation.section.embedded.ruby"]
expect(tokens[3]).toEqual value: '%Q-', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","meta.embedded.line.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.begin.ruby"] expect(tokens[3]).toEqual value: '%Q-', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","source.ruby.embedded.source","string.quoted.other.literal.upper.ruby","punctuation.definition.string.begin.ruby"]
expect(tokens[4]).toEqual value: 'crazy ideas', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","meta.embedded.line.ruby","string.quoted.other.literal.upper.ruby"] expect(tokens[4]).toEqual value: 'crazy ideas', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","source.ruby.embedded.source","string.quoted.other.literal.upper.ruby"]
expect(tokens[5]).toEqual value: '-', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","meta.embedded.line.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.end.ruby"] expect(tokens[5]).toEqual value: '-', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","source.ruby.embedded.source","string.quoted.other.literal.upper.ruby","punctuation.definition.string.end.ruby"]
expect(tokens[6]).toEqual value: '}', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","meta.embedded.line.ruby","punctuation.section.embedded.end.ruby","source.ruby"] # I'm not sure why it ends with source.ruby, but that is how the the ruby syntax specifies it expect(tokens[6]).toEqual value: '}', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","source.ruby.embedded.source","punctuation.section.embedded.ruby"]
expect(tokens[7]).toEqual value: ' for ruby syntax', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby"] expect(tokens[7]).toEqual value: ' for ruby syntax', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby"]
expect(tokens[8]).toEqual value: '+', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.end.ruby"] expect(tokens[8]).toEqual value: '+', scopes: ["source.ruby","string.quoted.other.literal.upper.ruby","punctuation.definition.string.end.ruby"]
expect(tokens[9]).toEqual value: ' ', scopes: ["source.ruby"] expect(tokens[9]).toEqual value: ' ', scopes: ["source.ruby"]