I don't think it would be just enough to add check in clearActiveFormattingElements. nodeStack.pop getting called all over the parser. I think we need fix on both ends: add nil return on empty stack pop and check in callers to act appropriately

// pop pops the stack. It will panic if s is empty.
func (s *nodeStack) pop() *Node {
	i := len(*s)
        if i == 0{
            return nil
        }
	n := (*s)[i-1]
	*s = (*s)[:i-1]
	return n
}

sry writing from phone....

I believe there is a reason for pop being documented to panic, probably performance.

In the original code if len(p.afe) == 0 was already there, it was just missing the initial check. I'd be more inclined to do some spot fixes where needed, but this is just my idea to address it. I'm open to suggestions.

cc @bradfitz

We don't seem to have a specific owner for x/net/html (on https://dev.golang.org/owners), so if you want to take any particular approach to a fix I doubt anyone will object.

(@mikioh is listed as the primary owner for the x/net repo overall, but @nigeltao and @namusyaka seem to have made most of the changes in the html package.)

The x/net/html package implements the HTML5 parsing algorithm, which has a very, very detailed specification:

https://html.spec.whatwg.org/multipage/syntax.html#tree-construction

Please bear this in mind. I'd like to avoid making ad-hoc algorithmic changes if possible.

CC'ing @TomAnthony who has also recently made a fix to x/net/html.

I only have familiarity with aspects of the algorithm, but my understanding is...

The pop shouldn't be empty, (the final call to pop leaving the open elements empty should be popping the html tag) and thus hitting the panic implies an error earlier in the tree construction.

I think in this case the the issue is with this part of the spec:

Where the steps above say to close the cell, they mean to run the following algorithm:

1. Generate implied end tags.

2. If the current node is not now a td element or a th element, then this is a parse error.

3. Pop elements from the stack of open elements stack until a td element or a th element has been popped from the stack.

4. Clear the list of active formatting elements up to the last marker.

5. Switch the insertion mode to "in row".

The relevant code is at line 1690:

if !p.elementInScope(tableScope, p.tok.DataAtom) {
	// Ignore the token.
	return true
}
// Close the cell and reprocess.
p.popUntil(tableScope, a.Td, a.Th)
p.clearActiveFormattingElements()
p.im = inRowIM

The check in step 2 of the spec is never completed, and in this case would be a failure. Then at step 3 we don't do anything (but we shouldn't be there according to the failure at step 2). My initial idea is that we combine these steps as popUntil reports whether it made any changes:

if !p.elementInScope(tableScope, p.tok.DataAtom) {
	// Ignore the token.
	return true
}
// Close the cell and reprocess.
if (p.popUntil(tableScope, a.Td, a.Th)) {
	p.clearActiveFormattingElements()
}
p.im = inRowIM

This passes the case above (and all tests) and produces the same output as both Chrome and Firefox for this snippet.

If this looks good I can prepare a PR with this and add a test case.

I've just come back to this. In the spec there is a similar situation in body insertion mode (see An end tag token whose tag name is one of: "applet", "marquee", "object"):

1. Generate implied end tags.

2. If the current node is not an HTML element with the same tag name as that of the token, then this is a parse error.

3. Pop elements from the stack of open elements until an HTML element with the same tag name as the token has been popped from the stack.

4. Clear the list of active formatting elements up to the last marker.

This is handled in the code at line 1127, with a construct that is the same as my suggestion above:

if p.popUntil(defaultScope, p.tok.DataAtom) {
	p.clearActiveFormattingElements()
}

Change https://golang.org/cl/172377 mentions this issue: html: Add missing condition to 'in cell' insertion mode, required by spec