On Sunday, 3 August 2014 at 23:51:30 UTC, Mike Farnsworth wrote:
This all seems to have a very simple solution, to use something
like: expect()
GCC for example has an intrinsic, __builtin_expect() that is
used to notify the compiler of a data constraint it can use in
optimization for branches. Why not make something like this a
first-class citizen in D (and even expand the concept to more
than just branch prediction)?
That way you don't have to hijack the meaning of assert(), but
optimizations can be made based on the condition.
__buitin_expect() in gcc usually just figures the expected
condition is fulfilled the vast majority of the time, but it
could be expanded to make a lack of fulfillment trigger an
exception (in non-release mode). And the compiler is always
free to optimize with the assumption the expectation is met.
Indeed, having a new function instead of hijacking assert would
seem to be the obvious solution.
That's really interesting about the possibility of conveying
probabilistic information to the compiler. Of course, we'd need
different functions for constant axioms and probabilistic ones:
we could use (for example) assume() for constants and expect()
for probabilities.