Bugs, limitations, and questionable design choices
==================================================

All known divergence from baseline (CLL+xorlo), or from plausible
interpretations thereof, is noted here.

Bugs:
    Question words other than {xu},{ma} and {mo} are not handled properly.

    The parser is only an approximation of camxes, resulting in some bugs:
	free clauses (vocatives, {to} brackets, {sei} etc) sometimes lead to
	    parse errors if they're not at the start of a (sub)sentence;
	{zoi} quotes don't work properly;
	magic words aren't handled accurately, and erasure cmavo not at all.

Limitations:
    Most indicators and frees are ignored, even those such as "irrealis"
    attitudinals which arguably have logical meaning.

    {ko} is not handled specially.

    {ra'o} is ignored.

    We parse the name in "LA sumtiTail", whereas arguably we should just use
    the text. Similarly, we don't keep hold of the raw text of a lu-quote.

    We don't handle {nei} or {no'a}, which have murky and self-referential
    semantics.

    We don't handle {soi}.

Deliberate deviations from baseline, not considered bugs:
    {na broda} is equivalent to {naku broda}; consequently, {ja'a} has no
    logical effect. This is in line with the BPFK section "brivla negators".

    Quantifiers don't scope over (even connected) sentence boundaries, unless
    prenexed.

    We consider {PA da} to introduce a quantifier in the usual logical sense,
    so ignore CLL's rules on rebinding bound variables (CLL:16.14) and
    simultaneous quantification in termsets (CLL:16.7) which conflict with
    this.

    Some constructs are accepted which the official grammar rejects; e.g. the
    tag grammar we use is essentially that of xorxes' zasni gerna.

Other behaviour whose correctness may not be immediately obvious:
    The {vo'a} series can only refer backwards. To see why, consider
	{vo'e broda ko'a gi'e brode ko'e}.