Overload Jinja2 autoescape for (La)TeX - markup

Question

Is it possible to overload Jinja2's autoescape so that it escapes something in a user-specified way (i.e. something other than HTML such as LaTeX)?
Here's an example trying to escape TeX.
import jinja2
class MyEnv(jinja2.Environment):
def __init__(self, filters={}, globals={}, tests={},
loader=None, extensions=[], **kwargs):
super(MyEnv, self).__init__(
autoescape = True,
)
template = MyEnv().from_string("""\documentclass[{{ class }}]
\\begin{document}
{{ content }}
\end{document}
""")
print template.render({
'class':'memoir',
'content': '{bob} <-- is escaped',
})
When you run the above, it outputs:
\documentclass[memoir]
\begin{document}
{bob} <-- is escaped
\end{document}
The problem here is that HTML escaping is used. So { and } should be escaped, but they're not, and < is converted to < but it should not be.
I'd like to overload the escape function that Jinja2 uses to escape variables.
My first thought is to overload finalize and disable autoescape. e.g.
import jinja2
class MyEnv(jinja2.Environment):
def __init__(self, filters={}, globals={}, tests={},
loader=None, extensions=[], **kwargs):
super(MyEnv, self).__init__(
autoescape = False, # turn off autoescape
finalize = self.finalize,
)
def finalize(self, s):
import re
if isinstance(s, jinja2.Markup):
return s
s = s.replace('\\', '')
s = s.replace('~', '\\textasciitilde')
s = re.sub(r'([#|^|$|&|%|{|}])', r'\\\1', s)
s = re.sub(r'_', r'\\_', s)
return jinja2.Markup(s)
template = MyEnv().from_string("""\documentclass[{{ class }}]
\\begin{document}
{{ content }}
\end{document}
""")
print template.render({
'class':'memoir',
'content': '{bob} <-- is escaped',
})
The output is incorrect, because the main text isn't made into Markup (i.e. a string flagged as safe):
documentclass[memoir]
begin\{document\}
\{bob\} <-- is escaped
end\{document\}
If I set autoescape to True, and leave in finalize it almost works (and in this example, it does work):
\documentclass[memoir]
\begin{document}
\{bob\} <-- is escaped
\end{document}
Turning autoescape on works because it makes the main body of text for the template as Markup (i.e. safe).
However, here's where the problem lies, if I change the input to a list that's joined:
template = MyEnv().from_string("""\documentclass[{{ class }}]
\\begin{document}
{{ content|join(" > a & b > "|safe) }}
\end{document}
""")
print template.render({
'class':'memoir',
'content': ['A&B', 'C<D'],
})
When I run this I get:
\documentclass[memoir]
\begin{document}
A&B > a & b > C<D
\end{document}
It would seem HTML autoescape is being run on the elements of 'content', rather than finalize. The simplest solution, provided Jinja2 and its autoescaping are loosely coupled, would seem to be to overload a autoescape function. I can't seem to figure that out, and the best I've come up with is the finalize function.
Is there a better way to handle escaping of TeX than overloading the finalize function? Can one overload autoescape?
For example, could one install a custom Markup package? (a choice I'd prefer to avoid)
Thank you for reading.