TagSoup is pretty poor — it's far from being an HTML parser (as defined by the HTML spec). The big problem is that because it's really just a tokenizer, it doesn't do any of the handling of mis-matched tags, which is really the interesting part of parsing HTML. Even insofar as what HTML can be parsed by a streaming parser, it's not great, as it doesn't imply any tags (which are a big part of HTML!).

It does, in its own code, state:

    -- We make some generalisations:
    -- <!name is a valid tag start closed by >
    -- <?name is a valid tag start closed by ?>
    -- </!name> is a valid closing tag
    -- </?name> is a valid closing tag
    -- <a "foo"> is a valid tag attibute in ! and ?, i.e missing an attribute name
    -- We also don't do lowercase conversion
    -- Entities are handled without a list of known entity names
    -- We don't have RCData, CData or Escape modes (only effects dat and tagOpen)

All of which mean that it doesn't actually follow what the HTML spec says! Just because it claims to support the spec doesn't mean much!