diff options
| author | Robin Haberkorn <rhaberkorn@fmsbw.de> | 2026-11-18 00:59:34 +0100 |
|---|---|---|
| committer | Robin Haberkorn <rhaberkorn@fmsbw.de> | 2026-11-18 00:59:34 +0100 |
| commit | 45ef41a78b8dd0564b9d704ce2c4ad3c1b983ae5 (patch) | |
| tree | 903774e7beeb359c8afc848f97e9d57ae3515ca1 /src/interface-gtk/fallback.css | |
| parent | 40b7968a3d7435a80678a169a1e811431ff7a7ad (diff) | |
more consistent integer and floating point arithmetics without undefined behaviormaster-fmsbw-ci
* `-1^*0` is 1, just like `1^*0`
* Floating point arithmetics use inf and NaN instead of errors.
In particular `1./0` and `1.^/0` no longer throw errors.
We better stay with IEEE 754 semantics than striving for consistency
with integer arithmetics. Lua does the same with its hybrid integer/float
type system.
The check for 0 was flawed anyway.
* C has lots of undefined behavior with regard to shifts, so we
normalize them just like Lua does: Negative displacements shift in the
other direction. A displacement >= 64 is always 0.
* Integers are allowed to overflow silently. The overflow behavior is
undefined in C, but since we assume two's complement representation,
they will actually be defined in practice.
This is also what Lua does.
Alternatively we could use checked additions, subtractions and
multiplications to catch all overflows and throw errors.
* There should be no undefined behavior in arithmetics in SciTECO.
Diffstat (limited to 'src/interface-gtk/fallback.css')
0 files changed, 0 insertions, 0 deletions
