stand-alone * or _ lead to incorrect results

[only-dev-time]

A text with several stand-alone underscores is not converted correctly.

input

**_ _ _ N _ _ _ _ _ _ _ R S _ _ N _**

output

 <p><strong><em> _ </em> N _ <em> _ </em> _ <em> _ R S _ </em> N _</strong></p>

expected

<p><strong>_ _ _ N _ _ _ _ _ _ _ R S _ _ N _</strong></p>

It seems that underscores with the surrounding spaces are replaced by the placeholder.
In the original on https://daringfireball.net/projects/markdown/dingus it is converted correctly.

[waylan]

I'm curious if the behavior changed here in #792 (version 3.2) or #629 (version 3.0). In other words, did we introduce this behavior in either of those changes or has it always been this way? @facelessuser you are the most familiar with the current cade as you made those changes. Any thoughts? I won't have time to take a closer look until next week at the earliest.

[facelessuser]

This has nothing to do with the surrounding **, but is a question of what should _ _ N _ _ _ _ R S _ _ N _ be parsed as.

#629 did not cause this, I am certain of that.

Could #805 (the fix for #792) have caused this? It is tough to tell without knowing what it was doing prior to this. No tests were changed in that review, only new ones added. I'm skeptical whether this changed the behavior, but I'd have to check.

_ and * are handled in two separate passes, they've never had direct awareness that if * was parsed on the outside that _ should be ignored within. _ can be parsed within words if "smart" is disabled.

Anyways, the first task would be to see if we broke any previous behavior.

If we did not break existing behavior, then this would be a new request. Then we would need to determine specifically what rule we are breaking (if any), and if so, see if we could add the necessary intelligence.

[facelessuser]

As far as #805 (the fix for #792) being the cause, I see no evidence. I reverted the changes related to * and _ locally. The first is the reverted, the second is the tip of master. Beahvior is identical

➜  markdown git:(master) ✗ python3
Python 3.11.0 (v3.11.0:deaf509e8f, Oct 24 2022, 14:43:23) [Clang 13.0.0 (clang-1300.0.29.30)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import markdown
>>> markdown.markdown('**_ _ _ N _ _ _ _ _ _ _ R S _ _ N _**')
'<p><strong><em> _ </em> N _ <em> _ </em> _ <em> _ R S _ </em> N _</strong></p>'
>>> exit()
➜  markdown git:(master) ✗ python3
Python 3.11.0 (v3.11.0:deaf509e8f, Oct 24 2022, 14:43:23) [Clang 13.0.0 (clang-1300.0.29.30)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import markdown
>>> markdown.markdown('**_ _ _ N _ _ _ _ _ _ _ R S _ _ N _**')
'<p><strong><em> _ </em> N _ <em> _ </em> _ <em> _ R S _ </em> N _</strong></p>'

[facelessuser]

It is highly doubtful that #629 is the culprit either. The only extensions that were drastically changed were the link patterns. The bold and emphasis extensions began using the new processor, but their fundamental patterns did not change, nor any sequencing of the extensions.

[facelessuser]

So bold and italic are handled in separate passes as they've always been done. At no point as Python Markdown ever not parsed _ simply because they were surrounded by **. If this was desired, we'd have to come up with a new mechanism to allow that. I will leave that up to @waylan to consider.

[waylan]

I finally had some time to look at this. It appears that we are ignoring this provision of the rules:

But if you surround an * or _ with spaces, it’ll be treated as a literal asterisk or underscore.

And that is a bug. That is also why I wondered if we had introduced a change in any of the more recent changes. Regardless of how the current behavior was introduced, we do need to fix this.

That said, the fix is not exactly clear. For example, the first and last underscore in the example input are not 'surrounded' by spaces as there is no space between the asterisk and the underscore in each instance. However, there are clearly instances where underscores which are surrounded by spaces are not being treated as literal underscores in the example.

To further complicate matters, we do sometimes properly ignore underscores surrounded by spaces. Consider this simple input:

Foo _ N _ Bar

... which properly renders as...

<p>Foo _ N _ Bar</p>

I haven't looked at the code to try to understand what is going on exactly, but something is amiss.

[facelessuser]

This is not a problem with my betterem package:

>>> markdown.markdown('**_ _ _ N _ _ _ _ _ _ _ R S _ _ N _**', extensions=['pymdownx.betterem'])
'<p><strong>_ _ _ N _ _ _ _ _ _ _ R S _ _ N _</strong></p>'

I know in my package, this is done by rejecting a * or _ as a starting token if it is immediately followed by white space and rejecting * or _ as an ending token if it is immediately preceded by white space.

[only-dev-time]

To further complicate matters, we do sometimes properly ignore underscores surrounded by spaces. Consider this simple input:

Foo _ N _ Bar

I only checked the code once. It may sound a bit cheeky, but a single underscore is no problem.
Only if there are several in a row does the result get wrong.

Foo _ _ N _ _ bar

renders as

<p>Foo _ <em> N _ </em> bar</p>

1. Thought:

I wonder why in the handler of the SimpleTextInlineProcessor match group 1 (_) is replaced by a placeholder (this processor is obviously only used with the stand alone _ or *).
If only the underscore (match group 3) were replaced, this correct result would be rendered:

<p>Foo _ _ N _ _ bar</p>

However, this modification gives this result for the example above:

<p><strong><em> _ _ N _ _ _ _ _ _ _ R S _ _ N </em></strong></p>

Now the first and last underscore after the double asterisk is interpreted as a markdown sign, although this does not represent the beginning and end of a text.
Therefore my

2. Thought:

The original code renders

**_ N _**

to

<p><strong><em> N </em></strong></p>

which is also wrong and leads me to suspect that something might be wrong with the joined **_ and _** and __* and *__ and ___ and ***.
Perhaps @facelessuser's method would be helpful. However, I cannot find and test the extension right now.

[facelessuser]

The extension is found in this extension pack: https://pypi.org/project/pymdown-extensions/.

Keep in mind, it may not match all corner cases if compared to some other parser, but it does aim to parse strong and italic cases in as sane a way as possible using the current methodology that Python Markdown employs.

[waylan]

I know in my package, this is done by rejecting a * or _ as a starting token if it is immediately followed by white space and rejecting * or _ as an ending token if it is immediately preceded by white space.

Yeah, this is how most implementations actually work in practice. In fact, the Commonmark spec codifies this behavior in great detail. However, the original rules explicitly state that the character should be rejected as a token if it 'surrounded' by spaces. Of course, the rule is vague and leaves out a bunch of edge cases (white space other than spaces, punctuation, etc).

That said, we can't get too hung up on the original rules as we explicitly don't follow them anyway with underscores to avoid middle-word emphasis. In the end we should probably strive to follow the behavior of the reference implementation with asterisks and copy that behavior to underscores minus middle-word emphasis.