A branch label is not a use of the pool constant with its number #648
Loading…
Reference in a new issue
No description provided.
Delete branch "rpg-9-an-label-not-a-use-20261028"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
A branch label is not a use of the pool constant with its number
The evaluator writes a float constant that no instruction uses as its own
literal as the decimal literal it twins (
3.0, not3.0E0). A decimalconstant that an instruction names is a literal of its own, so it is no twin,
and
namedfound those instructions by comparing every operand with thesymbol number. Operand 0 of the branch operations 15 to 18 is a label number,
and labels count on their own. In a large program the label numbers run into
the pool symbol numbers, so a branch label made a decimal constant look
named, the twin was refused, and the float was written with an exponent.
The held-out exam showed it: three float constants of the rebuilt object stood
in the source-order block of the pool where the reference holds the decimal
literals, so the rebuilt object had two more symbols than the reference.
namednow skips operand 0 of operations 15 to 18.A float operation beside a decimal literal also needs its converted operand
marked. The reference of statement 179 converts a packed field to float (
91)and adds a float constant (
47); the evaluator wroteB + 1.0, which thecompiler computes in decimal (
54). A converted operand beside a decimaltwin literal now takes
%FLOAT(...). A float literal with an exponent alreadymakes the operation float and gets none.
ftwin4-j1andftwin4-j2are thepublic reproduction: their emit differed from the twin in instructions, control
flow and the creation data, and now round-trips to
match;ftwin4-j3, theliteral with an exponent, matched before and still does.
Gates