Paginating through hits is a problem when hits are broken up by other tags.

[jan-niestadt]

If a hit e.g. begins in one sentence and ends in another, BLS' /docs/DOCID/contents will highlight it in two parts, and the frontend will think those are two separate hits.

For example, if we highlight a hit spanning from pancakes to They're (inclusive), BLS will return:

<s>I like <hl>pancakes.</hl></s>
<s><hl>They're</hl> delicious.</s>

So the individual hits cannot always be found from BLS' highlighted document contents. But the frontend's pagination uses the <hl> tags as if they represent individual hits, which breaks in this case, showing two parts of a hit as if there's two separate hits.

Another problem is overlapping hits.

Both BLS and frontend need to be changed to address these problems.
BLS should add a hit index attribute to any <hl> tag, so the frontend know the two tags belong together, e.g.:

<s>I like <hl n="1">pancakes.</hl></s>
<s><hl n="1">They're</hl> delicious.</s>

For two overlapping hits "The fox jumps" and "jumps over the dog":

<s><hl n="1">The fox <hl n="2">jumps</hl></hl><hl n="2"> over the dog.</hl></s>

The frontend would then use these indexes to identify and highlight whole hits at a time.

[KCMertens]

Also adding the hit start as an attribute would be useful. There are instances where we need to jump to a specific hit without necessarily knowing its index. For example when opening the document through the expanded snippet.