Chained summation operations like "=1.5+2^52-2^52" could benefit from Kahan algorithm implemented in tdf#137679. To do that, possibly we should change tokenizer (?), so that it treats several summation operations on the same level as one operation analogous to SUM. It should definitely not chain summations grouped by parentheses (I guess, it would be automatic, since operations inside parentheses are not "on the same level"). So e.g. "=(1.5+2^52)-2^52" would still return 2.
Summation of 2/3 terms won't really benefit much from Kahan. Let's hope there are not much extreme cases. tokenizer falls out of range of my knowledge. And also have to finish Kahan AVX once I finish my exams and some starmath related projects. Maybe it would be wise to modify the function wizard and specify in the description of SUM function that Kahan is being used. That they're different things and also notice in the documentation. Affectd file: https://opengrok.libreoffice.org/xref/core/sc/inc/scfuncs.hrc?r=24552a0b#955 /core/sc/inc/scfuncs.hrc:955