mirror of
https://github.com/wakatime/sublime-wakatime.git
synced 2023-08-10 21:13:02 +03:00
upgrade wakatime-cli to v8.0.3
This commit is contained in:
parent
6d4a4cf9eb
commit
fa40874635
@ -1,7 +1,7 @@
|
||||
__title__ = 'wakatime'
|
||||
__description__ = 'Common interface to the WakaTime api.'
|
||||
__url__ = 'https://github.com/wakatime/wakatime'
|
||||
__version_info__ = ('8', '0', '2')
|
||||
__version_info__ = ('8', '0', '3')
|
||||
__version__ = '.'.join(__version_info__)
|
||||
__author__ = 'Alan Hamlett'
|
||||
__author_email__ = 'alan@wakatime.com'
|
||||
|
@ -11,8 +11,8 @@
|
||||
|
||||
|
||||
LANGUAGES = {
|
||||
'typescript': 0.01,
|
||||
'f#': 0.01,
|
||||
'perl': 0.01,
|
||||
'perl6': 0.01,
|
||||
'f#': 0.01,
|
||||
'typescript': 0.01,
|
||||
}
|
||||
|
@ -134,9 +134,9 @@ class MatlabLexer(RegexLexer):
|
||||
}
|
||||
|
||||
def analyse_text(text):
|
||||
if re.match('^\s*%', text, re.M): # comment
|
||||
if re.search(r'^\s*%', text, re.M): # comment
|
||||
return 0.2
|
||||
elif re.match('^!\w+', text, re.M): # system cmd
|
||||
elif re.search(r'^!\w+', text, re.M): # system cmd
|
||||
return 0.2
|
||||
|
||||
|
||||
|
@ -172,10 +172,7 @@ def get_language_from_extension(file_name):
|
||||
if os.path.exists(u('{0}{1}').format(u(filepart), u('.c'))) or os.path.exists(u('{0}{1}').format(u(filepart), u('.C'))):
|
||||
return 'C'
|
||||
|
||||
directory = os.path.dirname(file_name)
|
||||
available_files = os.listdir(directory)
|
||||
available_extensions = list(zip(*map(os.path.splitext, available_files)))[1]
|
||||
available_extensions = [ext.lower() for ext in available_extensions]
|
||||
available_extensions = extensions_in_same_folder(file_name)
|
||||
if '.cpp' in available_extensions:
|
||||
return 'C++'
|
||||
if '.c' in available_extensions:
|
||||
@ -300,7 +297,7 @@ def custom_pygments_guess_lexer_for_filename(_fn, _text, **options):
|
||||
rv = lexer.analyse_text(_text)
|
||||
if rv == 1.0:
|
||||
return lexer(**options)
|
||||
result.append((rv, customize_priority(lexer)))
|
||||
result.append(customize_lexer_priority(_fn, rv, lexer))
|
||||
|
||||
def type_sort(t):
|
||||
# sort by:
|
||||
@ -308,16 +305,33 @@ def custom_pygments_guess_lexer_for_filename(_fn, _text, **options):
|
||||
# - is primary filename pattern?
|
||||
# - priority
|
||||
# - last resort: class name
|
||||
return (t[0], primary[t[1]], t[1].priority, t[1].__name__)
|
||||
return (t[0], primary[t[2]], t[1], t[2].__name__)
|
||||
result.sort(key=type_sort)
|
||||
|
||||
return result[-1][1](**options)
|
||||
return result[-1][2](**options)
|
||||
|
||||
|
||||
def customize_priority(lexer):
|
||||
"""Return an integer priority for the given lexer object."""
|
||||
def customize_lexer_priority(file_name, accuracy, lexer):
|
||||
"""Customize lexer priority"""
|
||||
|
||||
priority = lexer.priority
|
||||
|
||||
lexer_name = lexer.name.lower().replace('sharp', '#')
|
||||
if lexer_name in LANGUAGES:
|
||||
lexer.priority = LANGUAGES[lexer_name]
|
||||
return lexer
|
||||
priority = LANGUAGES[lexer_name]
|
||||
elif lexer_name == 'matlab':
|
||||
available_extensions = extensions_in_same_folder(file_name)
|
||||
if '.mat' in available_extensions:
|
||||
priority = 0.06
|
||||
|
||||
return (accuracy, priority, lexer)
|
||||
|
||||
|
||||
def extensions_in_same_folder(file_name):
|
||||
"""Returns a list of file extensions from the same folder as file_name."""
|
||||
|
||||
directory = os.path.dirname(file_name)
|
||||
files = os.listdir(directory)
|
||||
extensions = list(zip(*map(os.path.splitext, files)))[1]
|
||||
extensions = set([ext.lower() for ext in extensions])
|
||||
return extensions
|
||||
|
Loading…
Reference in New Issue
Block a user