Output expected value on assertion

[jimmy-mcelwain]

In reference to #268.

I am going to make this a draft because I definitely want feedback and expect that I will have to make changes.

The second most recent commit (as of now), 6668a4d had a function for each assertion, but there was a lot of code duplication. 7c68007 changed that by using more macros.

[gavanderhoorn]

do we really need two specialised versions for this?

I'm sort-of hoping/expecting we can use token pasting / concatenation and just build the expected vs actual string part of the msg using the pre-processor. I don't know whether that's true though, I expect you've tried.

[jimmy-mcelwain]

The problem is the format specifier in the print statement. In this case, I could probably get rid of one, because I only have specialized versions for pointers and integers and it wouldn't be disastrous to print out a pointer with say %d (or an integer with %p). But if we add asserts for other non-integer types like floats that wouldn't print properly with %d then we would definitely need specialized versions.

[gavanderhoorn]

I'm wondering whether this is what we'd want to do. With the current implementation, we'd be adding both mpSetAlarm(..) and IO state management code to all call-sites of this macro.

It'd be invisible -- because it's part of the macro body -- but it would still be there. At the very least it increases binary size, but it also seems like we should be able to call a function to which we delegate this responsibility -- didn't we have that before?

[jimmy-mcelwain]

I can move some of the macro stuff into a function to reduce the binary size.

[gavanderhoorn]

This is a nice use of token pasting / concatenation, but I might suggest keeping the actual 'ops' (ie: the conditionals) in the macros that call this one.

That would both make those macros easier to read (as we just see them doing something like if (!(x > y)) assert(my_msg)) and it would reduce the responsibilities of this macro (as it now only needs to log the message and raise an alarm (but see my other comment about this)).

Less responsibilities would simplify the code (less nesting), which is always good I feel.

[gavanderhoorn]

I can't immediately find it, but isn't it possible to simplify things a bit by considering the conditional expression a token and evaluate it both as an actual expression as well as use it as part of the message?

So we could end up with something like: motoRos_Assert(x < y, "my message if not true", ...) and x < y would end up in the message as well, despite being an expression. That would reduce the need for specialised macros, while still getting us what we're after.

---

Edit: something like this:

#define ASSERT_WITH_MSG_(EXPR, msg) \
{ \
  if (!(EXPR)) \
  { \
    printf("ERROR: %s: '%s'", msg, #EXPR ); \
  } \
} while (0)

For ASSERT_WITH_MSG_(1 < 1, "not smaller than") this prints:

ERROR: not smaller than: '1 < 1'

---

Edit 2: see frameworks like nemequ/munit fi, which do something similar (here fi).

---

Edit 3: ah, but for expressions which reference variables you'd only get the name, not the value: ASSERT_WITH_MSG_(some_struct.field0 < 3, "...") only prints:

ERROR: ...: 'some_struct.field0 < 3'

[jimmy-mcelwain]

Yeah, I think that the 2 layers are sort of needed if we want to

1. Reduce code duplication
2. Print out really detailed messages

I can get rid of the second macro call if we give up on one of these two.

[jimmy-mcelwain]

I made some small changes that should at least reduce binary size a little bit, nothing big. I'll have to test it more, but I'm mostly looking for design suggestions. I am open to changes, but this has also gotten a bit more complicated than I was hoping and I'm no macro expert.