[hartmath] How are evaluation rules specified?

[Preston P. <pr...@dw...>]

I've been delving into the means by which functions' evaluation rules
are specified.  I expect to need to extend these rules if I add to the
system, and it's also a core aspect that I should understand.  Here's
what I've found.

In symbolic evaluation, Session.EV() calls Session.ELoop() repeatedly,
and ELoop calls HObject.evaluate() on objects from the current
expression.  HFunction.evaluate() delegates the task to
HSymbol.evaluateFunction(), since the symbol is what stores the
information about the type of function.  evaluateFunction can apply "up"
or "down" rules, which actually cause changes.  For instance, this is an
applied "down" rule:
  D(Pow(t,5),t) becomes Multiply(5,Pow(t,4))

And this is where it gets awkward (apparently with reason).  Rules for
built-in objects are specified as long arrays of hard-coded longs and
bytes, which are passed to Session.setRules() to be evaluated.  It
appears that this difficult-to-examine technique is used to decrease
the time required to initialize the rules when they are loaded on the
fly. (correct?)

I'm curious as to how these arrays of numbers were generated.  It
seems as though they were constructed in a human-readable syntax,
converted into this format with some tool, and then block-copied into
the Java files in com.hartmath.loadable.*.  Is that correct?  If so:

  a) was the human-readable source syntax for these rules preserved?
  b) what tool performed the transformation?

I would like it if we could get to the point where the original form of
these rules were stored in CVS.  (is it now?)  Ideally, I would prefer
that an automated system update the hard-coded arrays whenever the
human-readable rules were modified, but that's probably more trouble
than it's worth.  I'd be happy with a test that confirms that the
human-readable and hard-coded forms are consistent.  If the original
source of the rules has been lost, it may be possible to use the
conversion tool to regenerate it.  I'd be glad to work on that.

I'd appreciate any confirmation, contradiction, or explanation of the
things I've described above.  Thanks.

==
Preston Pfarner
pf...@us...