On Thu 03 Jan 2019 10:42:30 PM CET, Eric Blake wrote: > In my view, code generators make sense when used on code that is > expected to change over time (a good example is QAPI because we add > new commands every release; other places might be a generator to help > deal with syscall handlers since newer kernels can add a syscall; or > even the fact that we have used some powerful GNU make textual > processing to make it easy to add files to particular subsets of the > build with as few lines edited as possible), where the goal is that > the generator gives you both a compact representation that is easier > to edit, and that the expansion from the generator ensures that > repetitive boilerplate is formed without typos. In short, if a > generator results in a net reduction in lines of edited source in > relation to the lines it produces, AND if the source that gets > regenerated is likely to change, then it makes total sense to spend > time on the generator. But when the amount of effort to write a > generator costs as much as just hard-coding the list outright, > especially when the list is not going to change (there really aren't > any other powers of 2 within 64 bits), I'm not sure a generator adds > any value.
I agree with Eric. Berto