28. Some Implementation Details

soelim (a standard Unix utility) is used to construct the parser flex and bison source files from scan.so, parse.so, and the numeric and range functions defined in the nf/ and rf/ subdirectories. Temporary files used for compilation are created in a tmp/ subdirectory.

In the doc/ subdirectory, soelim is also used to construct the documentation from SS.so and temporary files created by running SS to obtain the example outputs and lists of operators, functions, etc. The documentation is processed using sdf (Simple Document Format) to create a single HTML file as well as separate files for each section.

The internal data structures include: Cell, containing row and column numbers and flags for the row or column being fixed or relative; Range, a set of two Cells indicating the start and end of a rectangular region of the spreadsheet; Symbol, containing the name and formula of a symbol; and Node, which can hold a Cell, Range, Symbol, numeric value, string, operator, function code, or pointers to children Nodes in a binary tree. Nodes also contain a next pointer used to form linked-lists of expressions, such as function argument lists.

All operators and functions are stored in a table func indexed by the operator character value or function code. This makes it easy to evaluate operators and functions, i.e. in func.c eval_tree() for Node t:

  r = func[t->type]->f( t, ref_cell);

Numeric vs. Range functions

All of the numeric functions take a fixed number of arguments; most of the range functions take an indefinite list of arguments, but some of them (dot, feval, and llsq) take a fixed number of arguments. The essential difference between numeric and range functions is that arguments which are ranges are expanded into lists of cells for a numeric function, whereas for a range function they are left as ranges. Internally this difference makes it easier to evaluate numeric functions because they never have to traverse a range.

Range return values for numeric functions are also expanded into lists of cells. For range functions, range return values may or may not be expanded into lists of cells, depending on the function; for stats they are expanded, but for llsq they are not (since llsq needs to know the size of the coefficient return value range).

The following example demonstrates that the range argument of pow (a numeric function) is expanded into a list of cells, whereas the range arguments of stats, llsq, and sum (range functions) are left as ranges. The return values of frexp and stats are expanded into lists of cells, whereas the return value of llsq is not.

% cat numeric_vs_range.ss

a0 = pow(b0:c0); {a1:a4} = stats(b0:c0); a5 = sum(b0:c0);
{a6,a7:a8} = llsq("poly",b0:c0,b1:c1); {a9:a10} = frexp(sqrt(2));
print symbols formulas;

% ss < numeric_vs_range.ss

  $1 = {A1,A2,A3,A4} = stats(B0:C0) = 0
  $2 = {A6,A7:A8} = llsq("poly",B0:C0,B1:C1) = 0
  $3 = {A9,A10} = frexp(sqrt(2)) = 0
	A	B	C
0	pow(B0,C0)
1	($1)
2	($1)
3	($1)
4	($1)
5	sum(B0:C0)
6	($2)
7	($2)
8	($2)
9	($3)
10	($3)