Any progress on this? Having to go back and clean up things like http://en.wikipedia.org/w/index.php?title=Talk:Will.i.am&diff=646794855&oldid=646794596 is getting old...

Will investigate this tomorrow.

This seems a bug in our HTML output that emits a bare ":" at start of line as in this output below.

[subbu@earth lib] echo ":<span>a\n::b" | node parse --fetchConfig false --normalize
[warning][enwiki/Main Page] DSR inconsistency: cs/s mismatch for node: SPAN s: 7 ; cs: 8

<dl>
<dd><span>a</span></dd>
</dl>
<p>:b</p>

Normally selser would have let this through, but as of https://gerrit.wikimedia.org/r/#/c/180032/ we now examine all lines of output for any missing nowiki escapes because of edits elsewhere on the line. But, I think we are being overeager here about how we apply wt-escaping logic to selser output. So, there are two issues here:

• Problem with our wt2html output in handling unclosed tags which is throwing off our list building logic
• Problem with our wikitext escaping logic and how we apply it.

Looks like both of these issues identified above are regressions caused by other fixes.

• @Arlolra looks like 220374b36ac70f100333050cffdfc6de8d51876e broke this edge case. The fix needs to be sensitive to newlines.
• @cscott let us discuss how to handle the selser / nowiki scenario with the chunk-based serializer on IRC.

To clarify, I am not sure we should fix the selser / nowiki issue necessarily. The nowiki insertion is sane and correct and our old selser behavior was basically helping us by hiding parser bugs that let through wikitext constructs unparsed.

Change 192615 had a related patch set uploaded (by Arlolra):
Open tags only affect line when parsing definition list colon

https://gerrit.wikimedia.org/r/192615

Patch-For-Review

Change 192615 merged by jenkins-bot:
Open tags only affect line when parsing definition list colon

https://gerrit.wikimedia.org/r/192615

The patch gives us,

λ (master) echo -e ":<span>a\n::b" | node parse --fetchConfig false --normalize

<dl>
<dd><span>a
<dl>
<dd>b</dd>
</dl>
</span></dd>
</dl>

What was the conclusion of the nowiki discussion? Doesn't sound like you wanted to do anything, so safe to close?

I don't see how the chunk-based serializer is related here? Unless we think we could use the chunk-based serializer to be a little smarter about serialization?

It is related because it is now examining every line for nowiki escaping, even if there were no edits to that section of the document. The earlier serializer did not. While I can argue that it is probably a good thing to do by exposing parsing bugs, it also has the potential of introducing nowikis in unedited sections of the document. So, I was trying to find out how you approached this.