rgbds

rgbds copied to clipboard

• Foo * 0 is always 0
• Foo & 0 is always 0
• Foo - Foo is always 0
• Foo ^ Foo is always 0

These could be constant even when Foo is floating and unaligned.

• Foo + ~Foo and Foo | ~Foo are always -1
• Foo + -Foo is always 0
• Foo * 1 is always Foo; Foo * -1 is always -Foo (helps reducing into the other cases)

...and so on

Reducing those will prove very difficult, however. (And time-consuming if the comparison strategy is to compare the RPN buffers.)

I see what you mean, assuming Foo can be literally any expression, not just a symbol. In that case only expr * 0 and expr & 0 would be feasible.

And besides, I have considered those a long time ago, but I think they are too fringe to be worth special-casing. #965 made sense to be able to poll for alignment, but I can't see use cases for these.

And besides, I have considered those a long time ago, but I think they are too fringe to be worth special-casing. #965 made sense to be able to poll for alignment, but I can't see use cases for these.

The best use case I can think of is !cond * val0 + !!cond * val1 as a cheap inline conditional. (Yes, I know, true conditional expressions are coming someday, but someday is far away.)

Base the implementation for expr * 0 and expr & 0 on #976.

Closing as not worth the complexity.