Skip to content

numeric-conversion.md

Latest commit

 

History

History
92 lines (63 loc) · 3.63 KB

numeric-conversion.md

File metadata and controls

92 lines (63 loc) · 3.63 KB
Raw

How Do We Convert an Int to a Float?

First distinguish three meanings

  1. Change an NBT numeric tag's storage type.
  2. Enter a floating-point computation domain.
  3. Format a number as command syntax or text.

These are different source-language operations and should not share one lowering.

Scoreboard integer to NBT float

Use the target type and scale of execute store:

execute store result storage mdl:runtime out float 1 run scoreboard players get #value mdl.reg

For a fixed-point integer where 1000 represents 1.0:

execute store result storage mdl:runtime out float 0.001 run scoreboard players get #value mdl.reg

This is the preferred boundary conversion candidate: one ordinary pre-parsed command, no macro.

NBT number to NBT float through the result channel

Use data get as the command result and store it using the target numeric type:

execute store result storage mdl:runtime out float 1 run data get storage mdl:runtime input

This path is lossy. The 26.2 probes establish that data get multiplies and floors, then saturates into signed i32: 1.9 → 1, -1.9 → -2, 1.25 × 10 → 12, and 1e30 → INT_MAX. The final store converts that integer result to the requested NBT type. Use data modify ... set from instead when the source is already the desired native numeric type and the value must be copied without this projection.

Macro numeric retyping

Because numeric macro arguments are inserted without their original NBT suffix, a template can potentially attach a new suffix:

$data modify storage mdl:runtime out set value $(value)f

Status: Measured on 26.2. Double→float and float→double suffix templates worked, including expected binary32 narrowing at 16777217. For a simple score conversion, the ordinary execute store form remains the better pre-parsed baseline.

Do not implement general Float arithmetic in NBT

NBT stores floats but does not provide a general arithmetic instruction set. Moving every addition through NBT would create expensive command sequences.

Recommendation:

  • represent computation-heavy real numbers as fixed-point scoreboard integers;
  • carry a scale in the static type, such as Fixed<1000>;
  • perform add/subtract and carefully scaled multiply/divide in scores;
  • convert to NBT float/double only at Minecraft command boundaries;
  • use a true NBT float when the value is primarily transported, not computed.

The type checker can prevent mixing incompatible scales, and range analysis can detect possible 32-bit scoreboard overflow.

Constant conversions

Compile-time constants should be converted by the compiler and emitted in the final required syntax. They need no runtime command.

Measured and remaining tests

  • Measured: negative values, zero, float/double precision boundaries, scale, floor/saturation, macro suffix conversion, underflow, overflow, NaN, and both infinities produced through display transformations.
  • Remaining: every numeric target width, subnormal/tie boundaries, performance, fixed-point multiply/divide algorithms, and an explicit source-language rounding and non-finite-value policy.

The full observations and representation alternatives are in ranges/floating-point-semantics.md and ranges/floating-point-representations.md.

Sources