mirror of
https://github.com/httpie/cli.git
synced 2024-11-28 08:38:44 +02:00
182 lines
5.0 KiB
Python
182 lines
5.0 KiB
Python
"""
|
|
Colorizing of HTTP messages and content processing.
|
|
|
|
"""
|
|
import os
|
|
import re
|
|
import json
|
|
import pygments
|
|
from pygments import token, lexer
|
|
from pygments.styles import get_style_by_name, STYLE_MAP
|
|
from pygments.lexers import get_lexer_for_mimetype
|
|
from pygments.formatters.terminal import TerminalFormatter
|
|
from pygments.formatters.terminal256 import Terminal256Formatter
|
|
from pygments.util import ClassNotFound
|
|
from requests.compat import is_windows
|
|
from . import solarized
|
|
|
|
|
|
DEFAULT_STYLE = 'solarized'
|
|
AVAILABLE_STYLES = [DEFAULT_STYLE] + list(STYLE_MAP.keys())
|
|
if is_windows:
|
|
import colorama
|
|
colorama.init()
|
|
# 256 looks better on Windows
|
|
|
|
|
|
class HTTPLexer(lexer.RegexLexer):
|
|
"""
|
|
Simplified HTTP lexer for Pygments.
|
|
|
|
It only operates on headers and provides a stronger contrast between
|
|
their names and values than the original one bundled with Pygments
|
|
(`pygments.lexers.text import HttpLexer`), especially when
|
|
Solarized color scheme is used.
|
|
|
|
"""
|
|
name = 'HTTP'
|
|
aliases = ['http']
|
|
filenames = ['*.http']
|
|
tokens = {
|
|
'root': [
|
|
|
|
# Request-Line
|
|
(r'([A-Z]+)( +)([^ ]+)( +)(HTTP)(/)(\d+\.\d+)',
|
|
lexer.bygroups(
|
|
token.Name.Function,
|
|
token.Text,
|
|
token.Name.Namespace,
|
|
token.Text,
|
|
token.Keyword.Reserved,
|
|
token.Operator,
|
|
token.Number
|
|
)),
|
|
|
|
# Response Status-Line
|
|
(r'(HTTP)(/)(\d+\.\d+)( +)(\d{3})( +)(.+)',
|
|
lexer.bygroups(
|
|
token.Keyword.Reserved, # 'HTTP'
|
|
token.Operator, # '/'
|
|
token.Number, # Version
|
|
token.Text,
|
|
token.Number, # Status code
|
|
token.Text,
|
|
token.Name.Exception, # Reason
|
|
)),
|
|
|
|
# Header
|
|
(r'(.*?)( *)(:)( *)(.+)', lexer.bygroups(
|
|
token.Name.Attribute, # Name
|
|
token.Text,
|
|
token.Operator, # Colon
|
|
token.Text,
|
|
token.String # Value
|
|
))
|
|
]}
|
|
|
|
|
|
class BaseProcessor(object):
|
|
|
|
enabled = True
|
|
|
|
def __init__(self, env, **kwargs):
|
|
self.env = env
|
|
self.kwargs = kwargs
|
|
|
|
def process_headers(self, headers):
|
|
return headers
|
|
|
|
def process_body(self, content, content_type):
|
|
return content
|
|
|
|
|
|
class JSONProcessor(BaseProcessor):
|
|
|
|
def process_body(self, content, content_type):
|
|
if content_type == 'application/json':
|
|
try:
|
|
# Indent and sort the JSON data.
|
|
content = json.dumps(
|
|
json.loads(content),
|
|
sort_keys=True,
|
|
ensure_ascii=False,
|
|
indent=4,
|
|
)
|
|
except ValueError:
|
|
# Invalid JSON - we don't care.
|
|
pass
|
|
return content
|
|
|
|
|
|
class PygmentsProcessor(BaseProcessor):
|
|
|
|
def __init__(self, *args, **kwargs):
|
|
super(PygmentsProcessor, self).__init__(*args, **kwargs)
|
|
|
|
if not self.env.colors:
|
|
self.enabled = False
|
|
return
|
|
|
|
try:
|
|
style = get_style_by_name(
|
|
self.kwargs.get('pygments_style', DEFAULT_STYLE))
|
|
except ClassNotFound:
|
|
style = solarized.SolarizedStyle
|
|
|
|
if is_windows or self.env.colors == 256:
|
|
fmt_class = Terminal256Formatter
|
|
else:
|
|
fmt_class = TerminalFormatter
|
|
self.formatter = fmt_class(style=style)
|
|
|
|
def process_headers(self, headers):
|
|
return pygments.highlight(
|
|
headers, HTTPLexer(), self.formatter)
|
|
|
|
def process_body(self, content, content_type):
|
|
try:
|
|
lexer = get_lexer_for_mimetype(content_type)
|
|
except ClassNotFound:
|
|
pass
|
|
else:
|
|
content = pygments.highlight(content, lexer, self.formatter)
|
|
return content
|
|
|
|
|
|
class OutputProcessor(object):
|
|
|
|
installed_processors = [
|
|
JSONProcessor,
|
|
PygmentsProcessor
|
|
]
|
|
|
|
def __init__(self, env, **kwargs):
|
|
self.env = env
|
|
processors = [
|
|
cls(env, **kwargs)
|
|
for cls in self.installed_processors
|
|
]
|
|
self.processors = [p for p in processors if p.enabled]
|
|
|
|
def process_headers(self, headers):
|
|
for processor in self.processors:
|
|
headers = processor.process_headers(headers)
|
|
return headers
|
|
|
|
def process_body(self, content, content_type):
|
|
content_type = content_type.split(';')[0]
|
|
|
|
application_match = re.match(
|
|
r'application/(.+\+)(json|xml)$',
|
|
content_type
|
|
)
|
|
if application_match:
|
|
# Strip vendor and extensions from Content-Type
|
|
vendor, extension = application_match.groups()
|
|
content_type = content_type.replace(vendor, '')
|
|
|
|
for processor in self.processors:
|
|
content = processor.process_body(content, content_type)
|
|
|
|
return content
|