Size coding is on topic; I agree with Peter Taylor that an objective winning criterion can be in theory formed. However, it doesn't seem to have many advantages over code-golf questions, which as @mınxomaτ pointed out are commonly answered in machine code.
Here are some issues that may come up:
Interpreted languages are automatically excluded, and languages without short binaries won't be competitve. We generally discourage questions with language restrictions.
- Should competitors be forced to use a single language/compiler?
What counts as machine code anyway? Should bytecode be allowed?
Optimizing an answer for size will require repeatedly making changes to the source, then recompiling and measuring the binary size. No one wants to do that.
Of course, there's another way to write answers for sizecoding challenges, which leads to a problem:
In conclusion, although it is not off-topic, sizecoding would eventually reduce to a more tedious form of code-golf in machine languages.