Removed alt
attribute from LaTeX rendered images.
The `alt` contains the LaTeX formula rendered as Unicode characters, and that may include non-standard characters that break the RSS feed encoding.
This commit is contained in:
parent
cfd929f36f
commit
261e5bfc0e
1 changed files with 46 additions and 27 deletions
|
@ -31,11 +31,11 @@ def call(*args, **kwargs):
|
|||
|
||||
|
||||
# Defines our basic inline image
|
||||
img_expr = '<img class="latex inline math-%s" alt="%s" id="%s" src="data:image/png;base64,%s">'
|
||||
img_expr = '<img class="latex inline math-%s" id="%s" src="data:image/png;base64,%s">'
|
||||
|
||||
# Defines multiline expression image
|
||||
multiline_img_expr = '''<div class="multiline-wrapper">
|
||||
<img class="latex multiline math-%s" alt="%s" id="%s" src="data:image/png;base64,%s"></div>'''
|
||||
multiline_img_expr = """<div class="multiline-wrapper">
|
||||
<img class="latex multiline math-%s" id="%s" src="data:image/png;base64,%s"></div>"""
|
||||
|
||||
# Base CSS template
|
||||
img_css = """<style scoped>
|
||||
|
@ -55,8 +55,8 @@ img.latex.inline {
|
|||
</style>"""
|
||||
|
||||
# Cache and temp file paths
|
||||
tmpdir = tempfile.gettempdir() + '/markdown-latex'
|
||||
cache_file = tmpdir + '/latex.cache'
|
||||
tmpdir = tempfile.gettempdir() + "/markdown-latex"
|
||||
cache_file = tmpdir + "/latex.cache"
|
||||
|
||||
|
||||
class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
||||
|
@ -75,16 +75,20 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
"""
|
||||
|
||||
# Math TeX extraction regex
|
||||
math_extract_regex = re.compile(r'(.+?)((\\\(.+?\\\))|(\$\$\n.+?\n\$\$\n))(.*)', re.MULTILINE | re.DOTALL)
|
||||
math_extract_regex = re.compile(
|
||||
r"(.+?)((\\\(.+?\\\))|(\$\$\n.+?\n\$\$\n))(.*)", re.MULTILINE | re.DOTALL
|
||||
)
|
||||
|
||||
# Math TeX matching regex
|
||||
math_match_regex = re.compile(r'\s*(\\\(.+?\\\))|(\$\$\n.+?\n\$\$\n)\s*', re.MULTILINE | re.DOTALL)
|
||||
math_match_regex = re.compile(
|
||||
r"\s*(\\\(.+?\\\))|(\$\$\n.+?\n\$\$\n)\s*", re.MULTILINE | re.DOTALL
|
||||
)
|
||||
|
||||
def __init__(self, *_, **__):
|
||||
if not os.path.isdir(tmpdir):
|
||||
os.makedirs(tmpdir)
|
||||
try:
|
||||
with open(cache_file, 'r') as f:
|
||||
with open(cache_file, "r") as f:
|
||||
self.cached = json.load(f)
|
||||
except (IOError, json.JSONDecodeError):
|
||||
self.cached = {}
|
||||
|
@ -94,7 +98,8 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
("dvipng", "args"): "-q -T tight -bg Transparent -z 9 -D 200",
|
||||
("delimiters", "text"): "%",
|
||||
("delimiters", "math"): "$",
|
||||
("delimiters", "preamble"): "%%"}
|
||||
("delimiters", "preamble"): "%%",
|
||||
}
|
||||
|
||||
def _latex_to_base64(self, tex):
|
||||
"""Generates a base64 representation of TeX string"""
|
||||
|
@ -104,18 +109,24 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
with os.fdopen(tmp_file_fd, "w") as tmp_file:
|
||||
tmp_file.write(self.tex_preamble)
|
||||
tmp_file.write(tex)
|
||||
tmp_file.write('\n\\end{document}')
|
||||
tmp_file.write("\n\\end{document}")
|
||||
|
||||
# compile LaTeX document. A DVI file is created
|
||||
status = call(('latex -halt-on-error -output-directory={:s} {:s}'
|
||||
.format(tmpdir, path)).split(),
|
||||
stdout=PIPE, timeout=10)
|
||||
status = call(
|
||||
(
|
||||
"latex -halt-on-error -output-directory={:s} {:s}".format(tmpdir, path)
|
||||
).split(),
|
||||
stdout=PIPE,
|
||||
timeout=10,
|
||||
)
|
||||
|
||||
# clean up if the above failed
|
||||
if status:
|
||||
self._cleanup(path, err=True)
|
||||
raise Exception("Couldn't compile LaTeX document." +
|
||||
"Please read '%s.log' for more detail." % path)
|
||||
raise Exception(
|
||||
"Couldn't compile LaTeX document."
|
||||
+ "Please read '%s.log' for more detail." % path
|
||||
)
|
||||
|
||||
# Run dvipng on the generated DVI file. Use tight bounding box.
|
||||
# Magnification is set to 1200
|
||||
|
@ -129,8 +140,10 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
# clean up if we couldn't make the above work
|
||||
if status:
|
||||
self._cleanup(path, err=True)
|
||||
raise Exception("Couldn't convert LaTeX to image." +
|
||||
"Please read '%s.log' for more detail." % path)
|
||||
raise Exception(
|
||||
"Couldn't convert LaTeX to image."
|
||||
+ "Please read '%s.log' for more detail." % path
|
||||
)
|
||||
|
||||
# Read the png and encode the data
|
||||
try:
|
||||
|
@ -157,7 +170,7 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
def run(self, lines):
|
||||
"""Parses the actual page"""
|
||||
# Checks for the LaTeX header
|
||||
use_latex = any(line == '[//]: # (latex: 1)' for line in lines)
|
||||
use_latex = any(line == "[//]: # (latex: 1)" for line in lines)
|
||||
if not use_latex:
|
||||
return lines
|
||||
|
||||
|
@ -165,7 +178,9 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
page = "\n".join(lines)
|
||||
|
||||
# Adds a preamble mode
|
||||
self.tex_preamble += self.config[("general", "preamble")] + "\n\\begin{document}\n"
|
||||
self.tex_preamble += (
|
||||
self.config[("general", "preamble")] + "\n\\begin{document}\n"
|
||||
)
|
||||
|
||||
# Figure out our text strings and math-mode strings
|
||||
tex_expr = self.math_extract_regex.findall(page)
|
||||
|
@ -176,7 +191,7 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
|
||||
# Parse the expressions
|
||||
new_cache = {}
|
||||
new_page = ''
|
||||
new_page = ""
|
||||
n_multiline_expressions = 0
|
||||
|
||||
while page:
|
||||
|
@ -200,21 +215,25 @@ class LaTeXPreprocessor(markdown.preprocessors.Preprocessor):
|
|||
new_cache[tex_hash] = data
|
||||
|
||||
if is_multiline and n_multiline_expressions > 0:
|
||||
new_page += '</p>'
|
||||
new_page += (multiline_img_expr if is_multiline else img_expr) % ('true', expr, tex_hash, data)
|
||||
new_page += "</p>"
|
||||
new_page += (multiline_img_expr if is_multiline else img_expr) % (
|
||||
"true",
|
||||
tex_hash,
|
||||
data,
|
||||
)
|
||||
|
||||
if is_multiline:
|
||||
new_page += '<p>'
|
||||
new_page += "<p>"
|
||||
n_multiline_expressions += 1
|
||||
|
||||
page = m.group(5)
|
||||
|
||||
if n_multiline_expressions > 0:
|
||||
new_page += '</p>'
|
||||
new_page += "</p>"
|
||||
|
||||
# Cache our data
|
||||
self.cached.update(new_cache)
|
||||
with open(cache_file, 'w') as f:
|
||||
with open(cache_file, "w") as f:
|
||||
json.dump(self.cached, f)
|
||||
|
||||
# Make sure to re-split the lines
|
||||
|
@ -231,6 +250,6 @@ class MarkdownLatex(markdown.Extension):
|
|||
def extendMarkdown(self, md):
|
||||
md.preprocessors.register(
|
||||
LaTeXPreprocessor(self),
|
||||
'latex',
|
||||
"latex",
|
||||
1,
|
||||
)
|
||||
|
|
Loading…
Reference in a new issue