[TUHS] Re: undiagnosed pic error

[Douglas McIlroy <douglas.mcilroy@dartmouth.edu>]

There may be a simple generic way to correct pic's habit of accepting
any set of object modifiers on any object, but obeying only a
compatible subset.

Pic already collects a bit vector of modifier types attached to the
current object. If that were extended with a few more bits that
designate the object types, the size, B, of the bit vector would be
about 35--an easy fit in one 64-bit word. Then a BxB bit matrix could
record both modifier/modifier incompatibilities and object/modifier
incompatibilities. The collected bit vector needs to be tested against
the matrix once per object definition.

It seems to be harder to catch duplication of modifiers, requiring
extra code at all points where bits are set. Nevertheless, this kind
of error also merits detection.

Some questions

Does anybody think the issue is not worth addressing?

Is there a better scheme than that suggested above?

Is the scheme adequate? It would not, for example, catch a three-way
incompatibility that does not entail any pairwise incompatibility,
should such an incompatibility exist.

Any other thoughts?

Doug