mirror of
https://github.com/pulsar-edit/pulsar.git
synced 2024-11-10 10:17:11 +03:00
178 lines
7.9 KiB
CoffeeScript
178 lines
7.9 KiB
CoffeeScript
path = require 'path'
|
|
fs = require 'fs-plus'
|
|
temp = require('temp').track()
|
|
GrammarRegistry = require '../src/grammar-registry'
|
|
Grim = require 'grim'
|
|
|
|
describe "the `grammars` global", ->
|
|
beforeEach ->
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-text')
|
|
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-javascript')
|
|
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-coffee-script')
|
|
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-ruby')
|
|
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-git')
|
|
|
|
afterEach ->
|
|
atom.packages.deactivatePackages()
|
|
atom.packages.unloadPackages()
|
|
temp.cleanupSync()
|
|
|
|
describe ".selectGrammar(filePath)", ->
|
|
it "always returns a grammar", ->
|
|
registry = new GrammarRegistry(config: atom.config)
|
|
expect(registry.selectGrammar().scopeName).toBe 'text.plain.null-grammar'
|
|
|
|
it "selects the text.plain grammar over the null grammar", ->
|
|
expect(atom.grammars.selectGrammar('test.txt').scopeName).toBe 'text.plain'
|
|
|
|
it "selects a grammar based on the file path case insensitively", ->
|
|
expect(atom.grammars.selectGrammar('/tmp/source.coffee').scopeName).toBe 'source.coffee'
|
|
expect(atom.grammars.selectGrammar('/tmp/source.COFFEE').scopeName).toBe 'source.coffee'
|
|
|
|
describe "on Windows", ->
|
|
originalPlatform = null
|
|
|
|
beforeEach ->
|
|
originalPlatform = process.platform
|
|
Object.defineProperty process, 'platform', value: 'win32'
|
|
|
|
afterEach ->
|
|
Object.defineProperty process, 'platform', value: originalPlatform
|
|
|
|
it "normalizes back slashes to forward slashes when matching the fileTypes", ->
|
|
expect(atom.grammars.selectGrammar('something\\.git\\config').scopeName).toBe 'source.git-config'
|
|
|
|
it "can use the filePath to load the correct grammar based on the grammar's filetype", ->
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-git')
|
|
|
|
runs ->
|
|
expect(atom.grammars.selectGrammar("file.js").name).toBe "JavaScript" # based on extension (.js)
|
|
expect(atom.grammars.selectGrammar(path.join(temp.dir, '.git', 'config')).name).toBe "Git Config" # based on end of the path (.git/config)
|
|
expect(atom.grammars.selectGrammar("Rakefile").name).toBe "Ruby" # based on the file's basename (Rakefile)
|
|
expect(atom.grammars.selectGrammar("curb").name).toBe "Null Grammar"
|
|
expect(atom.grammars.selectGrammar("/hu.git/config").name).toBe "Null Grammar"
|
|
|
|
it "uses the filePath's shebang line if the grammar cannot be determined by the extension or basename", ->
|
|
filePath = require.resolve("./fixtures/shebang")
|
|
expect(atom.grammars.selectGrammar(filePath).name).toBe "Ruby"
|
|
|
|
it "uses the number of newlines in the first line regex to determine the number of lines to test against", ->
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage('language-property-list')
|
|
|
|
runs ->
|
|
fileContent = "first-line\n<html>"
|
|
expect(atom.grammars.selectGrammar("dummy.coffee", fileContent).name).toBe "CoffeeScript"
|
|
|
|
fileContent = '<?xml version="1.0" encoding="UTF-8"?>'
|
|
expect(atom.grammars.selectGrammar("grammar.tmLanguage", fileContent).name).toBe "Null Grammar"
|
|
|
|
fileContent += '\n<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">'
|
|
expect(atom.grammars.selectGrammar("grammar.tmLanguage", fileContent).name).toBe "Property List (XML)"
|
|
|
|
it "doesn't read the file when the file contents are specified", ->
|
|
filePath = require.resolve("./fixtures/shebang")
|
|
filePathContents = fs.readFileSync(filePath, 'utf8')
|
|
spyOn(fs, 'read').andCallThrough()
|
|
expect(atom.grammars.selectGrammar(filePath, filePathContents).name).toBe "Ruby"
|
|
expect(fs.read).not.toHaveBeenCalled()
|
|
|
|
describe "when multiple grammars have matching fileTypes", ->
|
|
it "selects the grammar with the longest fileType match", ->
|
|
grammarPath1 = temp.path(suffix: '.json')
|
|
fs.writeFileSync grammarPath1, JSON.stringify(
|
|
name: 'test1'
|
|
scopeName: 'source1'
|
|
fileTypes: ['test']
|
|
)
|
|
grammar1 = atom.grammars.loadGrammarSync(grammarPath1)
|
|
expect(atom.grammars.selectGrammar('more.test', '')).toBe grammar1
|
|
fs.removeSync(grammarPath1)
|
|
|
|
grammarPath2 = temp.path(suffix: '.json')
|
|
fs.writeFileSync grammarPath2, JSON.stringify(
|
|
name: 'test2'
|
|
scopeName: 'source2'
|
|
fileTypes: ['test', 'more.test']
|
|
)
|
|
grammar2 = atom.grammars.loadGrammarSync(grammarPath2)
|
|
expect(atom.grammars.selectGrammar('more.test', '')).toBe grammar2
|
|
fs.removeSync(grammarPath2)
|
|
|
|
it "favors non-bundled packages when breaking scoring ties", ->
|
|
waitsForPromise ->
|
|
atom.packages.activatePackage(path.join(__dirname, 'fixtures', 'packages', 'package-with-rb-filetype'))
|
|
|
|
runs ->
|
|
atom.grammars.grammarForScopeName('source.ruby').bundledPackage = true
|
|
atom.grammars.grammarForScopeName('test.rb').bundledPackage = false
|
|
|
|
expect(atom.grammars.selectGrammar('test.rb').scopeName).toBe 'test.rb'
|
|
|
|
describe "when there is no file path", ->
|
|
it "does not throw an exception (regression)", ->
|
|
expect(-> atom.grammars.selectGrammar(null, '#!/usr/bin/ruby')).not.toThrow()
|
|
expect(-> atom.grammars.selectGrammar(null, '')).not.toThrow()
|
|
expect(-> atom.grammars.selectGrammar(null, null)).not.toThrow()
|
|
|
|
describe "when the user has custom grammar file types", ->
|
|
it "considers the custom file types as well as those defined in the grammar", ->
|
|
atom.config.set('core.customFileTypes', 'source.ruby': ['Cheffile'])
|
|
expect(atom.grammars.selectGrammar('build/Cheffile', 'cookbook "postgres"').scopeName).toBe 'source.ruby'
|
|
|
|
it "favors user-defined file types over built-in ones of equal length", ->
|
|
atom.config.set('core.customFileTypes',
|
|
'source.coffee': ['Rakefile'],
|
|
'source.ruby': ['Cakefile']
|
|
)
|
|
expect(atom.grammars.selectGrammar('Rakefile', '').scopeName).toBe 'source.coffee'
|
|
expect(atom.grammars.selectGrammar('Cakefile', '').scopeName).toBe 'source.ruby'
|
|
|
|
it "favors user-defined file types over grammars with matching first-line-regexps", ->
|
|
atom.config.set('core.customFileTypes', 'source.ruby': ['bootstrap'])
|
|
expect(atom.grammars.selectGrammar('bootstrap', '#!/usr/bin/env node').scopeName).toBe 'source.ruby'
|
|
|
|
describe "when there is a grammar with a first line pattern, the file type of the file is known, but from a different grammar", ->
|
|
it "favors file type over the matching pattern", ->
|
|
expect(atom.grammars.selectGrammar('foo.rb', '#!/usr/bin/env node').scopeName).toBe 'source.ruby'
|
|
|
|
describe ".removeGrammar(grammar)", ->
|
|
it "removes the grammar, so it won't be returned by selectGrammar", ->
|
|
grammar = atom.grammars.selectGrammar('foo.js')
|
|
atom.grammars.removeGrammar(grammar)
|
|
expect(atom.grammars.selectGrammar('foo.js').name).not.toBe grammar.name
|
|
|
|
describe "grammar overrides", ->
|
|
it "logs deprecations and uses the TextEditorRegistry", ->
|
|
editor = null
|
|
|
|
waitsForPromise ->
|
|
atom.workspace.open('sample.js').then (e) -> editor = e
|
|
|
|
runs ->
|
|
spyOn(Grim, 'deprecate')
|
|
|
|
atom.grammars.setGrammarOverrideForPath(editor.getPath(), 'source.ruby')
|
|
expect(Grim.deprecate.callCount).toBe 1
|
|
expect(editor.getGrammar().name).toBe 'Ruby'
|
|
|
|
expect(atom.grammars.grammarOverrideForPath(editor.getPath())).toBe('source.ruby')
|
|
expect(Grim.deprecate.callCount).toBe 2
|
|
|
|
atom.grammars.clearGrammarOverrideForPath(editor.getPath(), 'source.ruby')
|
|
expect(Grim.deprecate.callCount).toBe 3
|
|
expect(editor.getGrammar().name).toBe 'JavaScript'
|
|
|
|
expect(atom.grammars.grammarOverrideForPath(editor.getPath())).toBe(undefined)
|
|
expect(Grim.deprecate.callCount).toBe 4
|