Interactive formula compiler and range estimator

Claim 2

A computer-implemented method for estimating the range of a numeric formula in an interactive computer program, comprising representing real numbers as intervals; breaking the formula into a sequence of primitive operations; estimating bounds for each primitive operation; converting decimal numbers precisely; displaying each interval in a cell; and evaluating said formula as a sequence of interval operations.

Claim 3

The method of claim 1 further comprising using the directed rounding operations of an IEEE floating point arithmetic processor to estimate said bounds.

Claim 4

The method of claim 1 further comprising optionally using pythagorean operations to estimate said bounds for sums or products.

Claim 5

The method of claim 1 further comprising translating formulas into processor commands.

Claim 6

The method of claim 1 further comprising optionally displaying only those digits where both interval endpoints agree.

Claim 7

A computer-implemented method for the tracking of numerical errors in an interactive computer program, comprising maintaining a collection of formulas in natural order; using quiet nan mantissa bits to track the source cell of an error, with higher order bits being used to classify the error type; evaluating said formulas with an IEEE floating point arithmetic processor; testing for an exception raised at the end of the evaluation of each said formulas; creating a quiet nan with a cell address encoding if said exception is raised; and displaying said nan with a pointer to the source of the error.

Claim 8

A computer-implemented method for the tracking of numerical errors in an interactive computer program, comprising maintaining a collection of formulas in natural order; using quiet nan mantissa bits to track the source cell of an error, with higher order bits being used to classify the error type; evaluating said formulas with an IEEE floating point arithmetic processor; testing for an exception raised at the end of the evaluation of each said formulas; creating a quiet nan with a cell address encoding if said exception is raised; and displaying said nan with a pointer to the source of the error.

Claim 9

The method of claim 6 further comprising using bits from said nan mantissa designating ERR and NA which are more significant than the address bits, so that all ERR values have mantissas numerically greater than those of all NA values.

Claim 10

The method of claim 6 further comprising converting infinities to ERR.

Claim 11

The method of claim 6 further comprising masking exceptions on said processor.

Claim 12

The method of claim 6 further comprising trapping exceptions for special error processing.

Claim 13

The method of claim 6 further comprising using signaling nans to represent strings or other nonnumeric data types.

Claim 14

A computer-implemented method for the evaluation of a collection of formulas in an interactive computer program, comprising representing said formulas in polish form; arranging said formulas in a natural order; translating said formulas into native code with inline floating point instructions; using infinities and nans for error propagation; anticipating coercions based on known cell and function return types; and using a function library for the more complicated functions.

Claim 15

A computer-implemented method for the evaluation of a collection of formulas in an interactive computer program, comprising representing said formulas in polish form; arranging said formulas in a natural order; translating said formulas into native code with inline floating point instructions; using infinities and nans for error propagation; anticipating coercions based on known cell and function return types; and using a function library for the more complicated functions.

Claim 16

The method of claim 12 further comprising resolving addresses based on assumptions about memory locations and data types, and invalidating said native code if those assumptions change.

Claim 17

The method of claim 12 further comprising processing interval data types.

Claim 18

The method of claim 12 further comprising using nan mantissa bits to track the type of an error.

Claim 19

The method of claim 12 further comprising using nan mantissa bits to track the source cell of an error.

Claim 20

The method of claim 12 further comprising using signaling nans to represent strings or other non-numeric data types.

Claim 21

The method of claim 12 further comprising falling back to a p-code interpreter in case said native code is unavailable or invalid.

Claim 22

The method of claim 12 further comprising executing said native code on an IEEE floating point processor with exceptions masked; using status flags on said processor to test for errors at the end of execution of said formulas; and selectively saving the status flags in order to detect errors in a subexpression.

Claim 23

The method of claim 12 further comprising trapping exceptions to simulate a deep stack of floating point registers or for special error processing.