I have to determine whether or not a solution certified on one particular OS (Linux distro A) will run exactly the same way on another OS (Linux distro B) just because Linux distro B is fully binary compatible with Linux distro A.
My concern is that the solution uses a lot of libraries, certified on Linux distro A only. Those libraries are making calculations using floating points. Users are afraid that if the solution runs on Linux distro B, it could lead in inaccuracy in decimal for those calculations. This inaccuracy multiplied by billions of transactions becomes a serious issue.
As far as I understand, the inaccuracy of decimal (or let's say the difference of result) between 2 binaries of the same library is induced by the fact that, at the compilation, the generated assembly instructions are not the same (or in different order).
So, as long as those libraries are not recompiled, I could expect the exact same results on Linux distro A and Linux distro B, right? Or am I missing something?