##// END OF EJS Templates
Aligned expanded sumarry box content better and moved comments as last column
Aligned expanded sumarry box content better and moved comments as last column

File last commit:

r3444:e5ce0962 default
r3639:7ee2e326 new-ui
Show More
codeblocks.py
786 lines | 29.8 KiB | text/x-python | PythonLexer
# -*- coding: utf-8 -*-
# Copyright (C) 2011-2019 RhodeCode GmbH
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License, version 3
# (only), as published by the Free Software Foundation.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
# This program is dual-licensed. If you wish to learn more about the
# RhodeCode Enterprise Edition, including its added features, Support services,
# and proprietary license terms, please see https://rhodecode.com/licenses/
import logging
import difflib
from itertools import groupby
from pygments import lex
from pygments.formatters.html import _get_ttype_class as pygment_token_class
from pygments.lexers.special import TextLexer, Token
from pygments.lexers import get_lexer_by_name
from pyramid import compat
from rhodecode.lib.helpers import (
get_lexer_for_filenode, html_escape, get_custom_lexer)
from rhodecode.lib.utils2 import AttributeDict, StrictAttributeDict, safe_unicode
from rhodecode.lib.vcs.nodes import FileNode
from rhodecode.lib.vcs.exceptions import VCSError, NodeDoesNotExistError
from rhodecode.lib.diff_match_patch import diff_match_patch
from rhodecode.lib.diffs import LimitedDiffContainer, DEL_FILENODE, BIN_FILENODE
plain_text_lexer = get_lexer_by_name(
'text', stripall=False, stripnl=False, ensurenl=False)
log = logging.getLogger(__name__)
def filenode_as_lines_tokens(filenode, lexer=None):
org_lexer = lexer
lexer = lexer or get_lexer_for_filenode(filenode)
log.debug('Generating file node pygment tokens for %s, %s, org_lexer:%s',
lexer, filenode, org_lexer)
content = filenode.content
tokens = tokenize_string(content, lexer)
lines = split_token_stream(tokens, content)
rv = list(lines)
return rv
def tokenize_string(content, lexer):
"""
Use pygments to tokenize some content based on a lexer
ensuring all original new lines and whitespace is preserved
"""
lexer.stripall = False
lexer.stripnl = False
lexer.ensurenl = False
if isinstance(lexer, TextLexer):
lexed = [(Token.Text, content)]
else:
lexed = lex(content, lexer)
for token_type, token_text in lexed:
yield pygment_token_class(token_type), token_text
def split_token_stream(tokens, content):
"""
Take a list of (TokenType, text) tuples and split them by a string
split_token_stream([(TEXT, 'some\ntext'), (TEXT, 'more\n')])
[(TEXT, 'some'), (TEXT, 'text'),
(TEXT, 'more'), (TEXT, 'text')]
"""
token_buffer = []
for token_class, token_text in tokens:
parts = token_text.split('\n')
for part in parts[:-1]:
token_buffer.append((token_class, part))
yield token_buffer
token_buffer = []
token_buffer.append((token_class, parts[-1]))
if token_buffer:
yield token_buffer
elif content:
# this is a special case, we have the content, but tokenization didn't produce
# any results. THis can happen if know file extensions like .css have some bogus
# unicode content without any newline characters
yield [(pygment_token_class(Token.Text), content)]
def filenode_as_annotated_lines_tokens(filenode):
"""
Take a file node and return a list of annotations => lines, if no annotation
is found, it will be None.
eg:
[
(annotation1, [
(1, line1_tokens_list),
(2, line2_tokens_list),
]),
(annotation2, [
(3, line1_tokens_list),
]),
(None, [
(4, line1_tokens_list),
]),
(annotation1, [
(5, line1_tokens_list),
(6, line2_tokens_list),
])
]
"""
commit_cache = {} # cache commit_getter lookups
def _get_annotation(commit_id, commit_getter):
if commit_id not in commit_cache:
commit_cache[commit_id] = commit_getter()
return commit_cache[commit_id]
annotation_lookup = {
line_no: _get_annotation(commit_id, commit_getter)
for line_no, commit_id, commit_getter, line_content
in filenode.annotate
}
annotations_lines = ((annotation_lookup.get(line_no), line_no, tokens)
for line_no, tokens
in enumerate(filenode_as_lines_tokens(filenode), 1))
grouped_annotations_lines = groupby(annotations_lines, lambda x: x[0])
for annotation, group in grouped_annotations_lines:
yield (
annotation, [(line_no, tokens)
for (_, line_no, tokens) in group]
)
def render_tokenstream(tokenstream):
result = []
for token_class, token_ops_texts in rollup_tokenstream(tokenstream):
if token_class:
result.append(u'<span class="%s">' % token_class)
else:
result.append(u'<span>')
for op_tag, token_text in token_ops_texts:
if op_tag:
result.append(u'<%s>' % op_tag)
escaped_text = html_escape(token_text)
# TODO: dan: investigate showing hidden characters like space/nl/tab
# escaped_text = escaped_text.replace(' ', '<sp> </sp>')
# escaped_text = escaped_text.replace('\n', '<nl>\n</nl>')
# escaped_text = escaped_text.replace('\t', '<tab>\t</tab>')
result.append(escaped_text)
if op_tag:
result.append(u'</%s>' % op_tag)
result.append(u'</span>')
html = ''.join(result)
return html
def rollup_tokenstream(tokenstream):
"""
Group a token stream of the format:
('class', 'op', 'text')
or
('class', 'text')
into
[('class1',
[('op1', 'text'),
('op2', 'text')]),
('class2',
[('op3', 'text')])]
This is used to get the minimal tags necessary when
rendering to html eg for a token stream ie.
<span class="A"><ins>he</ins>llo</span>
vs
<span class="A"><ins>he</ins></span><span class="A">llo</span>
If a 2 tuple is passed in, the output op will be an empty string.
eg:
>>> rollup_tokenstream([('classA', '', 'h'),
('classA', 'del', 'ell'),
('classA', '', 'o'),
('classB', '', ' '),
('classA', '', 'the'),
('classA', '', 're'),
])
[('classA', [('', 'h'), ('del', 'ell'), ('', 'o')],
('classB', [('', ' ')],
('classA', [('', 'there')]]
"""
if tokenstream and len(tokenstream[0]) == 2:
tokenstream = ((t[0], '', t[1]) for t in tokenstream)
result = []
for token_class, op_list in groupby(tokenstream, lambda t: t[0]):
ops = []
for token_op, token_text_list in groupby(op_list, lambda o: o[1]):
text_buffer = []
for t_class, t_op, t_text in token_text_list:
text_buffer.append(t_text)
ops.append((token_op, ''.join(text_buffer)))
result.append((token_class, ops))
return result
def tokens_diff(old_tokens, new_tokens, use_diff_match_patch=True):
"""
Converts a list of (token_class, token_text) tuples to a list of
(token_class, token_op, token_text) tuples where token_op is one of
('ins', 'del', '')
:param old_tokens: list of (token_class, token_text) tuples of old line
:param new_tokens: list of (token_class, token_text) tuples of new line
:param use_diff_match_patch: boolean, will use google's diff match patch
library which has options to 'smooth' out the character by character
differences making nicer ins/del blocks
"""
old_tokens_result = []
new_tokens_result = []
similarity = difflib.SequenceMatcher(None,
''.join(token_text for token_class, token_text in old_tokens),
''.join(token_text for token_class, token_text in new_tokens)
).ratio()
if similarity < 0.6: # return, the blocks are too different
for token_class, token_text in old_tokens:
old_tokens_result.append((token_class, '', token_text))
for token_class, token_text in new_tokens:
new_tokens_result.append((token_class, '', token_text))
return old_tokens_result, new_tokens_result, similarity
token_sequence_matcher = difflib.SequenceMatcher(None,
[x[1] for x in old_tokens],
[x[1] for x in new_tokens])
for tag, o1, o2, n1, n2 in token_sequence_matcher.get_opcodes():
# check the differences by token block types first to give a more
# nicer "block" level replacement vs character diffs
if tag == 'equal':
for token_class, token_text in old_tokens[o1:o2]:
old_tokens_result.append((token_class, '', token_text))
for token_class, token_text in new_tokens[n1:n2]:
new_tokens_result.append((token_class, '', token_text))
elif tag == 'delete':
for token_class, token_text in old_tokens[o1:o2]:
old_tokens_result.append((token_class, 'del', token_text))
elif tag == 'insert':
for token_class, token_text in new_tokens[n1:n2]:
new_tokens_result.append((token_class, 'ins', token_text))
elif tag == 'replace':
# if same type token blocks must be replaced, do a diff on the
# characters in the token blocks to show individual changes
old_char_tokens = []
new_char_tokens = []
for token_class, token_text in old_tokens[o1:o2]:
for char in token_text:
old_char_tokens.append((token_class, char))
for token_class, token_text in new_tokens[n1:n2]:
for char in token_text:
new_char_tokens.append((token_class, char))
old_string = ''.join([token_text for
token_class, token_text in old_char_tokens])
new_string = ''.join([token_text for
token_class, token_text in new_char_tokens])
char_sequence = difflib.SequenceMatcher(
None, old_string, new_string)
copcodes = char_sequence.get_opcodes()
obuffer, nbuffer = [], []
if use_diff_match_patch:
dmp = diff_match_patch()
dmp.Diff_EditCost = 11 # TODO: dan: extract this to a setting
reps = dmp.diff_main(old_string, new_string)
dmp.diff_cleanupEfficiency(reps)
a, b = 0, 0
for op, rep in reps:
l = len(rep)
if op == 0:
for i, c in enumerate(rep):
obuffer.append((old_char_tokens[a+i][0], '', c))
nbuffer.append((new_char_tokens[b+i][0], '', c))
a += l
b += l
elif op == -1:
for i, c in enumerate(rep):
obuffer.append((old_char_tokens[a+i][0], 'del', c))
a += l
elif op == 1:
for i, c in enumerate(rep):
nbuffer.append((new_char_tokens[b+i][0], 'ins', c))
b += l
else:
for ctag, co1, co2, cn1, cn2 in copcodes:
if ctag == 'equal':
for token_class, token_text in old_char_tokens[co1:co2]:
obuffer.append((token_class, '', token_text))
for token_class, token_text in new_char_tokens[cn1:cn2]:
nbuffer.append((token_class, '', token_text))
elif ctag == 'delete':
for token_class, token_text in old_char_tokens[co1:co2]:
obuffer.append((token_class, 'del', token_text))
elif ctag == 'insert':
for token_class, token_text in new_char_tokens[cn1:cn2]:
nbuffer.append((token_class, 'ins', token_text))
elif ctag == 'replace':
for token_class, token_text in old_char_tokens[co1:co2]:
obuffer.append((token_class, 'del', token_text))
for token_class, token_text in new_char_tokens[cn1:cn2]:
nbuffer.append((token_class, 'ins', token_text))
old_tokens_result.extend(obuffer)
new_tokens_result.extend(nbuffer)
return old_tokens_result, new_tokens_result, similarity
def diffset_node_getter(commit):
def get_node(fname):
try:
return commit.get_node(fname)
except NodeDoesNotExistError:
return None
return get_node
class DiffSet(object):
"""
An object for parsing the diff result from diffs.DiffProcessor and
adding highlighting, side by side/unified renderings and line diffs
"""
HL_REAL = 'REAL' # highlights using original file, slow
HL_FAST = 'FAST' # highlights using just the line, fast but not correct
# in the case of multiline code
HL_NONE = 'NONE' # no highlighting, fastest
def __init__(self, highlight_mode=HL_REAL, repo_name=None,
source_repo_name=None,
source_node_getter=lambda filename: None,
target_repo_name=None,
target_node_getter=lambda filename: None,
source_nodes=None, target_nodes=None,
# files over this size will use fast highlighting
max_file_size_limit=150 * 1024,
):
self.highlight_mode = highlight_mode
self.highlighted_filenodes = {}
self.source_node_getter = source_node_getter
self.target_node_getter = target_node_getter
self.source_nodes = source_nodes or {}
self.target_nodes = target_nodes or {}
self.repo_name = repo_name
self.target_repo_name = target_repo_name or repo_name
self.source_repo_name = source_repo_name or repo_name
self.max_file_size_limit = max_file_size_limit
def render_patchset(self, patchset, source_ref=None, target_ref=None):
diffset = AttributeDict(dict(
lines_added=0,
lines_deleted=0,
changed_files=0,
files=[],
file_stats={},
limited_diff=isinstance(patchset, LimitedDiffContainer),
repo_name=self.repo_name,
target_repo_name=self.target_repo_name,
source_repo_name=self.source_repo_name,
source_ref=source_ref,
target_ref=target_ref,
))
for patch in patchset:
diffset.file_stats[patch['filename']] = patch['stats']
filediff = self.render_patch(patch)
filediff.diffset = StrictAttributeDict(dict(
source_ref=diffset.source_ref,
target_ref=diffset.target_ref,
repo_name=diffset.repo_name,
source_repo_name=diffset.source_repo_name,
target_repo_name=diffset.target_repo_name,
))
diffset.files.append(filediff)
diffset.changed_files += 1
if not patch['stats']['binary']:
diffset.lines_added += patch['stats']['added']
diffset.lines_deleted += patch['stats']['deleted']
return diffset
_lexer_cache = {}
def _get_lexer_for_filename(self, filename, filenode=None):
# cached because we might need to call it twice for source/target
if filename not in self._lexer_cache:
if filenode:
lexer = filenode.lexer
extension = filenode.extension
else:
lexer = FileNode.get_lexer(filename=filename)
extension = filename.split('.')[-1]
lexer = get_custom_lexer(extension) or lexer
self._lexer_cache[filename] = lexer
return self._lexer_cache[filename]
def render_patch(self, patch):
log.debug('rendering diff for %r', patch['filename'])
source_filename = patch['original_filename']
target_filename = patch['filename']
source_lexer = plain_text_lexer
target_lexer = plain_text_lexer
if not patch['stats']['binary']:
node_hl_mode = self.HL_NONE if patch['chunks'] == [] else None
hl_mode = node_hl_mode or self.highlight_mode
if hl_mode == self.HL_REAL:
if (source_filename and patch['operation'] in ('D', 'M')
and source_filename not in self.source_nodes):
self.source_nodes[source_filename] = (
self.source_node_getter(source_filename))
if (target_filename and patch['operation'] in ('A', 'M')
and target_filename not in self.target_nodes):
self.target_nodes[target_filename] = (
self.target_node_getter(target_filename))
elif hl_mode == self.HL_FAST:
source_lexer = self._get_lexer_for_filename(source_filename)
target_lexer = self._get_lexer_for_filename(target_filename)
source_file = self.source_nodes.get(source_filename, source_filename)
target_file = self.target_nodes.get(target_filename, target_filename)
raw_id_uid = ''
if self.source_nodes.get(source_filename):
raw_id_uid = self.source_nodes[source_filename].commit.raw_id
if not raw_id_uid and self.target_nodes.get(target_filename):
# in case this is a new file we only have it in target
raw_id_uid = self.target_nodes[target_filename].commit.raw_id
source_filenode, target_filenode = None, None
# TODO: dan: FileNode.lexer works on the content of the file - which
# can be slow - issue #4289 explains a lexer clean up - which once
# done can allow caching a lexer for a filenode to avoid the file lookup
if isinstance(source_file, FileNode):
source_filenode = source_file
#source_lexer = source_file.lexer
source_lexer = self._get_lexer_for_filename(source_filename)
source_file.lexer = source_lexer
if isinstance(target_file, FileNode):
target_filenode = target_file
#target_lexer = target_file.lexer
target_lexer = self._get_lexer_for_filename(target_filename)
target_file.lexer = target_lexer
source_file_path, target_file_path = None, None
if source_filename != '/dev/null':
source_file_path = source_filename
if target_filename != '/dev/null':
target_file_path = target_filename
source_file_type = source_lexer.name
target_file_type = target_lexer.name
filediff = AttributeDict({
'source_file_path': source_file_path,
'target_file_path': target_file_path,
'source_filenode': source_filenode,
'target_filenode': target_filenode,
'source_file_type': target_file_type,
'target_file_type': source_file_type,
'patch': {'filename': patch['filename'], 'stats': patch['stats']},
'operation': patch['operation'],
'source_mode': patch['stats']['old_mode'],
'target_mode': patch['stats']['new_mode'],
'limited_diff': patch['is_limited_diff'],
'hunks': [],
'hunk_ops': None,
'diffset': self,
'raw_id': raw_id_uid,
})
file_chunks = patch['chunks'][1:]
for hunk in file_chunks:
hunkbit = self.parse_hunk(hunk, source_file, target_file)
hunkbit.source_file_path = source_file_path
hunkbit.target_file_path = target_file_path
filediff.hunks.append(hunkbit)
# Simulate hunk on OPS type line which doesn't really contain any diff
# this allows commenting on those
if not file_chunks:
actions = []
for op_id, op_text in filediff.patch['stats']['ops'].items():
if op_id == DEL_FILENODE:
actions.append(u'file was removed')
elif op_id == BIN_FILENODE:
actions.append(u'binary diff hidden')
else:
actions.append(safe_unicode(op_text))
action_line = u'NO CONTENT: ' + \
u', '.join(actions) or u'UNDEFINED_ACTION'
hunk_ops = {'source_length': 0, 'source_start': 0,
'lines': [
{'new_lineno': 0, 'old_lineno': 1,
'action': 'unmod-no-hl', 'line': action_line}
],
'section_header': u'', 'target_start': 1, 'target_length': 1}
hunkbit = self.parse_hunk(hunk_ops, source_file, target_file)
hunkbit.source_file_path = source_file_path
hunkbit.target_file_path = target_file_path
filediff.hunk_ops = hunkbit
return filediff
def parse_hunk(self, hunk, source_file, target_file):
result = AttributeDict(dict(
source_start=hunk['source_start'],
source_length=hunk['source_length'],
target_start=hunk['target_start'],
target_length=hunk['target_length'],
section_header=hunk['section_header'],
lines=[],
))
before, after = [], []
for line in hunk['lines']:
if line['action'] in ['unmod', 'unmod-no-hl']:
no_hl = line['action'] == 'unmod-no-hl'
result.lines.extend(
self.parse_lines(before, after, source_file, target_file, no_hl=no_hl))
after.append(line)
before.append(line)
elif line['action'] == 'add':
after.append(line)
elif line['action'] == 'del':
before.append(line)
elif line['action'] == 'old-no-nl':
before.append(line)
elif line['action'] == 'new-no-nl':
after.append(line)
all_actions = [x['action'] for x in after] + [x['action'] for x in before]
no_hl = {x for x in all_actions} == {'unmod-no-hl'}
result.lines.extend(
self.parse_lines(before, after, source_file, target_file, no_hl=no_hl))
# NOTE(marcink): we must keep list() call here so we can cache the result...
result.unified = list(self.as_unified(result.lines))
result.sideside = result.lines
return result
def parse_lines(self, before_lines, after_lines, source_file, target_file,
no_hl=False):
# TODO: dan: investigate doing the diff comparison and fast highlighting
# on the entire before and after buffered block lines rather than by
# line, this means we can get better 'fast' highlighting if the context
# allows it - eg.
# line 4: """
# line 5: this gets highlighted as a string
# line 6: """
lines = []
before_newline = AttributeDict()
after_newline = AttributeDict()
if before_lines and before_lines[-1]['action'] == 'old-no-nl':
before_newline_line = before_lines.pop(-1)
before_newline.content = '\n {}'.format(
render_tokenstream(
[(x[0], '', x[1])
for x in [('nonl', before_newline_line['line'])]]))
if after_lines and after_lines[-1]['action'] == 'new-no-nl':
after_newline_line = after_lines.pop(-1)
after_newline.content = '\n {}'.format(
render_tokenstream(
[(x[0], '', x[1])
for x in [('nonl', after_newline_line['line'])]]))
while before_lines or after_lines:
before, after = None, None
before_tokens, after_tokens = None, None
if before_lines:
before = before_lines.pop(0)
if after_lines:
after = after_lines.pop(0)
original = AttributeDict()
modified = AttributeDict()
if before:
if before['action'] == 'old-no-nl':
before_tokens = [('nonl', before['line'])]
else:
before_tokens = self.get_line_tokens(
line_text=before['line'], line_number=before['old_lineno'],
input_file=source_file, no_hl=no_hl)
original.lineno = before['old_lineno']
original.content = before['line']
original.action = self.action_to_op(before['action'])
original.get_comment_args = (
source_file, 'o', before['old_lineno'])
if after:
if after['action'] == 'new-no-nl':
after_tokens = [('nonl', after['line'])]
else:
after_tokens = self.get_line_tokens(
line_text=after['line'], line_number=after['new_lineno'],
input_file=target_file, no_hl=no_hl)
modified.lineno = after['new_lineno']
modified.content = after['line']
modified.action = self.action_to_op(after['action'])
modified.get_comment_args = (target_file, 'n', after['new_lineno'])
# diff the lines
if before_tokens and after_tokens:
o_tokens, m_tokens, similarity = tokens_diff(
before_tokens, after_tokens)
original.content = render_tokenstream(o_tokens)
modified.content = render_tokenstream(m_tokens)
elif before_tokens:
original.content = render_tokenstream(
[(x[0], '', x[1]) for x in before_tokens])
elif after_tokens:
modified.content = render_tokenstream(
[(x[0], '', x[1]) for x in after_tokens])
if not before_lines and before_newline:
original.content += before_newline.content
before_newline = None
if not after_lines and after_newline:
modified.content += after_newline.content
after_newline = None
lines.append(AttributeDict({
'original': original,
'modified': modified,
}))
return lines
def get_line_tokens(self, line_text, line_number, input_file=None, no_hl=False):
filenode = None
filename = None
if isinstance(input_file, compat.string_types):
filename = input_file
elif isinstance(input_file, FileNode):
filenode = input_file
filename = input_file.unicode_path
hl_mode = self.HL_NONE if no_hl else self.highlight_mode
if hl_mode == self.HL_REAL and filenode:
lexer = self._get_lexer_for_filename(filename)
file_size_allowed = input_file.size < self.max_file_size_limit
if line_number and file_size_allowed:
return self.get_tokenized_filenode_line(
input_file, line_number, lexer)
if hl_mode in (self.HL_REAL, self.HL_FAST) and filename:
lexer = self._get_lexer_for_filename(filename)
return list(tokenize_string(line_text, lexer))
return list(tokenize_string(line_text, plain_text_lexer))
def get_tokenized_filenode_line(self, filenode, line_number, lexer=None):
if filenode not in self.highlighted_filenodes:
tokenized_lines = filenode_as_lines_tokens(filenode, lexer)
self.highlighted_filenodes[filenode] = tokenized_lines
try:
return self.highlighted_filenodes[filenode][line_number - 1]
except Exception:
return [('', u'rhodecode diff rendering error')]
def action_to_op(self, action):
return {
'add': '+',
'del': '-',
'unmod': ' ',
'unmod-no-hl': ' ',
'old-no-nl': ' ',
'new-no-nl': ' ',
}.get(action, action)
def as_unified(self, lines):
"""
Return a generator that yields the lines of a diff in unified order
"""
def generator():
buf = []
for line in lines:
if buf and not line.original or line.original.action == ' ':
for b in buf:
yield b
buf = []
if line.original:
if line.original.action == ' ':
yield (line.original.lineno, line.modified.lineno,
line.original.action, line.original.content,
line.original.get_comment_args)
continue
if line.original.action == '-':
yield (line.original.lineno, None,
line.original.action, line.original.content,
line.original.get_comment_args)
if line.modified.action == '+':
buf.append((
None, line.modified.lineno,
line.modified.action, line.modified.content,
line.modified.get_comment_args))
continue
if line.modified:
yield (None, line.modified.lineno,
line.modified.action, line.modified.content,
line.modified.get_comment_args)
for b in buf:
yield b
return generator()