"""BlockParser divides non `.py` source files into blocks of code and markup text."""
import ast
import codecs
import re
from pathlib import Path
from textwrap import dedent
import pygments.lexers
import pygments.token
from sphinx.errors import ExtensionError
from sphinx.util.logging import getLogger
from .py_source_parser import FLAG_BODY, Block
logger = getLogger("sphinx-gallery")
# Don't just use "x in pygments.token.Comment" because it also includes preprocessor
# statements
COMMENT_TYPES = (
pygments.token.Comment.Single,
pygments.token.Comment,
pygments.token.Comment.Multiline,
)
[docs]
class BlockParser:
"""
A parser that breaks a source file into blocks of code and markup text.
Determines the source language and identifies comment blocks using pygments.
Parameters
----------
source_file : str
A file name that has a suffix compatible with files that are subsequently parsed
gallery_conf : dict
Contains the configuration of Sphinx-Gallery.
"""
def __init__(self, source_file, gallery_conf):
source_file = Path(source_file)
if name := gallery_conf["filetype_parsers"].get(source_file.suffix):
self.lexer = pygments.lexers.find_lexer_class_by_name(name)()
else:
self.lexer = pygments.lexers.find_lexer_class_for_filename(source_file)()
self.language = self.lexer.name
# determine valid comment syntaxes. For each possible syntax, the tuple contains
# - A test comment
# - The comment start character
# - The comment end character (for multiline comments) or None
# - A regex for the start of a block based on repeated characters, or None
comment_tests = [
("#= comment =#", "#=", "=#", None), # Julia multiline
("# comment", "#", None, "#{20,}"), # Julia, Ruby, Bash, Perl, etc.
("// comment", "//", None, "/{20,}"), # C++, C#, Java, Rust, etc.
("/* comment */", r"/\*", r"\*/", r"/\*{20,}/"), # C/C++ etc. multiline
("% comment", "%", None, "%{20,}"), # Matlab
("! comment", "!", None, "!{20,}"), # Fortran
("c comment", r"^c(?:$| )", None, None), # Fortran 77
]
self.allowed_comments = []
allowed_special = []
self.multiline_end = re.compile(chr(0)) # unmatchable regex
for test, start, end, special in comment_tests:
if next(self.lexer.get_tokens(test))[0] in COMMENT_TYPES:
self.allowed_comments.append(start)
if end:
self.multiline_end = re.compile(rf"(.*?)\s*{end}")
if special:
allowed_special.append(special)
if self.language == "Matlab":
# Matlab treats code sections starting with "%%" in a similar manner to
# tools that recognize "# %%" in Python code, so we accept that style as
# an alternative.
allowed_special.append(r"%%(?:$|\s)")
if r"/\*" in self.allowed_comments:
# Remove decorative asterisks and comment starts from C-style multiline
# comments
self.multiline_cleanup = re.compile(r"\s*/?\*\s*")
else:
self.multiline_cleanup = re.compile(r"\s*")
comment_start = "|".join(self.allowed_comments)
allowed_special = "|".join(allowed_special)
if allowed_special:
self.start_special = re.compile(
f"(?:(?:{comment_start}) ?%% ?|{allowed_special})(.*)"
)
else:
self.start_special = re.compile(f"(?:{comment_start}) ?%% ?(.*)")
self.continue_text = re.compile(f"(?:{comment_start}) ?(.*)")
# The pattern for in-file config comments is designed to not greedily match
# newlines at the start and end, except for one newline at the end. This
# ensures that the matched pattern can be removed from the code without
# changing the block structure; i.e. empty newlines are preserved, e.g. in
#
# a = 1
#
# # sphinx_gallery_thumbnail_number = 2
#
# b = 2
flag_start = rf"^[\ \t]*(?:{comment_start})\s*"
self.infile_config_pattern = re.compile(flag_start + FLAG_BODY, re.MULTILINE)
self.start_ignore_flag = flag_start + "sphinx_gallery_start_ignore"
self.end_ignore_flag = flag_start + "sphinx_gallery_end_ignore"
self.ignore_block_pattern = re.compile(
rf"{self.start_ignore_flag}(?:[\s\S]*?){self.end_ignore_flag}\n?",
re.MULTILINE,
)
[docs]
def split_code_and_text_blocks(self, source_file, return_node=False):
"""Return list with source file separated into code and text blocks.
Parameters
----------
source_file : str
Path to the source file.
return_node : bool
Ignored; returning an ast node is not supported
Returns
-------
file_conf : dict
File-specific settings given in source file comments as:
``# sphinx_gallery_<name> = <value>``
blocks : list
(label, content, line_number)
List where each element is a tuple with the label ('text' or 'code'),
the corresponding content string of block and the leading line number.
node : None
Returning an ast node is not supported.
"""
with codecs.open(source_file, "r", "utf-8") as fid:
content = fid.read()
# change from Windows format to UNIX for uniformity
content = content.replace("\r\n", "\n")
return self._split_content(content)
def _get_content_lines(self, content):
"""
Combine individual tokens into lines.
Use the first non-whitespace token (if any) as the characteristic token type for
the line.
"""
current_line = []
line_token = pygments.token.Whitespace
for token, text in self.lexer.get_tokens(content):
if line_token == pygments.token.Whitespace:
line_token = token
if "\n" in text:
text_lines = text.split("\n")
# first item belongs to the previous line
current_line.append(text_lines.pop(0))
yield line_token, "".join(current_line)
# last item belongs to the line after this token
current_line = [text_lines.pop()]
# Anything left is a block of lines to add directly
for ln in text_lines:
if not ln.strip():
line_token = pygments.token.Whitespace
yield line_token, ln
if not current_line[0].strip():
line_token = pygments.token.Whitespace
else:
current_line.append(text)
def _get_blocks(self, content):
"""
Generate a sequence of "blocks" from the lines in ``content``.
Each block is a tuple of (label, content, line_number), matching the format
ultimately returned by `split_code_and_text_blocks`.
"""
start_text = self.continue_text # No special delimiter needed for first block
needs_multiline_cleanup = False
def cleanup_multiline(lines):
nonlocal needs_multiline_cleanup
first_line = 1 if start_text == self.continue_text else 0
longest = max(len(line) for line in lines)
matched = False
for i, line in enumerate(lines[first_line:]):
if m := self.multiline_cleanup.match(line):
matched = True
if (n := len(m.group(0))) < len(line):
longest = min(longest, n)
if matched and longest:
for i, line in enumerate(lines[first_line:], start=first_line):
lines[i] = lines[i][longest:]
needs_multiline_cleanup = False
return lines
def finalize_block(mode, block):
nonlocal start_text
if mode == "text":
if needs_multiline_cleanup:
cleanup_multiline(block)
# subsequent blocks need to have the special delimiter
start_text = self.start_special
# Remove leading blank lines, and end in a single newline
first = 0
for i, line in enumerate(block):
first = i
if line.strip():
break
last = None
for i, line in enumerate(reversed(block)):
last = -i or None
if line.strip():
break
block.append("")
text = dedent("\n".join(block[first:last]))
else:
text = "\n".join(block)
return Block(mode, text, n - len(block))
block = []
mode = None
for n, (token, text) in enumerate(self._get_content_lines(content)):
if mode == "text" and token in pygments.token.Whitespace:
# Blank line ends current text block
if block:
yield finalize_block(mode, block)
mode, block = None, []
elif (
mode != "text"
and token in COMMENT_TYPES
and (m := start_text.search(text))
):
# start of a text block; end the current block
if block:
yield finalize_block(mode, block)
mode, block = "text", []
if (trailing_text := m.group(1)) is not None:
if start_text == self.continue_text:
if (delimited := self.start_special.search(text)) and (
heading := delimited.group(1).strip()
):
# Treat text on the same line as the delimiter as a title
block.extend((heading, "=" * len(heading), ""))
else:
# Keep any text on the first line of the title block as is
block.append(trailing_text)
elif heading := trailing_text.strip():
# Treat text on the same line as the delimiter as a heading
block.extend((heading, "-" * len(heading), ""))
elif mode == "text" and token in COMMENT_TYPES:
# Continuation of a text block
if m := self.start_special.search(text):
# Starting a new text block now is just a continuation of the
# existing one, possibly including a new heading.
if heading := m.group(1).strip():
block.extend((heading, "-" * len(heading), ""))
pass
elif token == pygments.token.Comment.Multiline:
if m := self.multiline_end.search(text):
block.append(m.group(1))
needs_multiline_cleanup = True
else:
block.append(text)
else:
block.append(self.continue_text.search(text).group(1))
elif mode != "code":
# start of a code block
if block:
yield finalize_block(mode, block)
mode, block = "code", [text]
else:
# continuation of a code block
block.append(text)
# end of input ends final block
if block:
yield finalize_block(mode, block)
def _split_content(self, content):
"""
Split the input content into blocks.
Return a tuple of (file_conf, blocks, None) that corresponds to the return
values of ``split_code_and_text_blocks``.
"""
file_conf = self.extract_file_config(content)
blocks = list(self._get_blocks(content))
# For examples that start with a code block before the file docstring due to
# language conventions or requirements, swap these blocks so the title block
# comes first and merge consecutive code blocks if needed.
if len(blocks) >= 2 and blocks[0].type == "code" and blocks[1].type == "text":
blocks[0], blocks[1] = blocks[1], blocks[0]
if len(blocks) >= 3 and blocks[2].type == "code":
blocks[1] = Block(
"code",
f"{blocks[1].content}\n{blocks[2].content}",
blocks[1].lineno,
)
blocks.pop(2)
return file_conf, blocks, None
[docs]
def remove_ignore_blocks(self, code_block):
"""
Return the content of *code_block* with ignored areas removed.
An ignore block starts with ``?? sphinx_gallery_start_ignore`` and ends with
``?? sphinx_gallery_end_ignore`` where ``??`` is the active language's line
comment marker. These lines and anything in between them will be removed, but
surrounding empty lines are preserved.
Parameters
----------
code_block : str
A code segment.
"""
num_start_flags = len(re.findall(self.start_ignore_flag, code_block))
num_end_flags = len(re.findall(self.end_ignore_flag, code_block))
if num_start_flags != num_end_flags:
raise ExtensionError(
'All "sphinx_gallery_start_ignore" flags must have a matching '
'"sphinx_gallery_end_ignore" flag!'
)
return re.subn(self.ignore_block_pattern, "", code_block)[0]