It is possible, but the proof's effort grows combinatorially with the size of the program. The large majority of "real software" is not amenable to this kind of test, making the mathematical theorem very concrete.
That you cannot have a program that will decide if arbitrary theoretical programs halt on arbitrary inputs has no bearing on how well specific real-world software is able to deal with real-world inputs. You can have a smartwatch, a virtual synthesizer and IoT switches that are guaranteed to have a fixed upper response time for the tasks they are designed for. That they don't provide these guarantees is only because most users prefer a device that works 99% of the time but is much cheaper.