Not thread safe

[vbraun]

I'm getting occasional not reproducable errors (with the same markdown source) when running on Appengine:

Traceback (most recent call last):
[...]
  File "/base/data/home/apps/.../deps/markdown/__init__.py", line 375, in convert
    newRoot = treeprocessor.run(root)
  File "/base/data/home/apps/.../deps/markdown/treeprocessors.py", line 330, in run
    self.ancestors.pop()
IndexError: pop from empty list

I'm having threading enabled (GAE default), and TreeProcessor has a global ancestors stack. So presumably this crashes because simultaneous threads run the markdown processor. I can't find anything in the docs whether python-markdown is supposed to be thread safe, so thats either a documentation request or a bug ;-)

[facelessuser]

Are running multiple threads on a single instance of the Markdown class? Or a separate instance in a separate thread? It is kind of unclear on what exactly you are doing.

With that said, there has been no effort to guarantee thread safety, especially when you consider that you can pull in any custom extension which may or may not also be thread safe.

Personally, I would not rely on Python Markdown being thread safe.

[waylan]

We have never put any work into testing or verifying if Python-Markdown is thread safe. Personally I have no idea if it is or not.

Curious, how are you calling Markdown? Do you call the markdown.markdown wrapper which creates a new class instance on each call, or are you creating a single class instance and then calling convert for each request? I would expect the former to be more thread safe than the later.

[vbraun]

Indeed, I have a Markdown singleton (with some custom extensions) that all threads access.

So to summarize:

• A Markdown instance has internal parse state and is not thread-safe
• Thread-local Markdown instances (e.g. instantiating it in the thread) is thread-safe

[waylan]

So to summarize:

• A Markdown instance has internal parse state and is not thread-safe
• Thread-local Markdown instances (e.g. instantiating it in the thread) is thread-safe

That is correct. I suppose we should document that.