Post #3433189
2026-06-27 19:04 UTC
@fclc@mast.hpc.social @fay59@tech.lgbt with some interval analysis and algebraic transformation, i would expect that most math operations in practice can be computed to the point you can detect overflow within 64 bits, maybe 128 once in a while
Replies (1)
-
@fay59@tech.lgbt 2026-06-27 19:14
@joe@f.duriansoftware.com @fclc@mast.hpc.social yeah, what I’m thinking is that at function boundaries you discourage AP integers, and you diagnose in places that the compiler cannot tell the upper bound of your integer. The diag is resolved by converting to a fixed-size integer type. Conversions that are known to succeed all the time Just Work, and if the compiler doesn’t know then you need to explicitly ask for truncation or an optional result