On Monday, 4 June 2018 at 15:48:35 UTC, DigitalDesigns wrote:
Does D have any methods of validating code in a natural manner
besides unit tests and contracts?
I'm specifically thinking of validating mathematical
calculations and boolean operations that could depend on very
improbable scenarios but are technically invalid logic.
These issues tend to creep up in calculations that involve
floating points due to various reasons or comparisons that
mistakenly use > for >= or vice versa.
If I have a variable such as a buffer which has a length and an
offset in to that buffer is calculated using double precision
then rounding errors could cause the offset to except the
length and create an access violation.
To be able to theoretically test all the possibilities all
valid inputs would need to be checked. One can setup unit tests
to test these possibilities but it can be difficult to cover
all cases in even a semi-complex program.
Just curious if something exists that allows for mathematical
validation such code in an relatively canonical way. This isn't
too hard for pure functions but dealing with non-pure functions
can be a pain.
Perhaps not quite what you're looking for, but I think you would
be interested in the LLVM fuzzing part of Johan Engelen's talk at
DConf 2018:
https://www.youtube.com/watch?v=GMKvYrjaaoU (at around 34:30).
