Even cpp would suffice, but for cpp we need to change the style of the comments.

Derived from #4

introduced in #49

introduced in #59

So it will cause the renaming of %label directive to %bind

which IMO makes more sense.

introduced in #49

introduced in #28

introduced in #37

introduced in #76

introduced in #18

In trying to encode the magic(5) definitions for the file(1) utility, I discovered that the magic marker 'BM' is already used here PC Bitmap.
It is documented Bitmap file header.

Perhaps the Birtual Machine magic marker could be redefined as БМ or as 'bm'.

introduced in #31

Derived from #4

introduced in #71

After #77 is implemented

introduced in #28

introduced in #31

On macos (no idea about "real" BSD), int64_t is defined as long long, while on linux it's defined as long. This causes issues with the printf formatting generating warnings (it won't actually affect the runtime since long is still also 8 bytes)

There's inttypes.h and PRId64 nonsense, but that might get ugly...

introduced in #18

introduced in #31

• Limit amount of iterations

• Print the generated numbers on each iteration

introduced in #31

Sample throw the lerp with regular intervals

introduced in #59

introduced in #59

The birtual machine file could have a marker, like "BRCH".

That way, the magic(5) database and file(1)command would tell the user that it is a Birtual Machine binary file.
That marker could be followed by a version, say a 32-bit.

Evidently, the bme would have to skip the two fields before executing any instructions.

Currently, mapping instructions to handlers is done using a switch statement. To my knowledge, because it's basically nested if statements, this will produce an O(log n) complexity mapping at best. You can use an array containing pointers to handlers, using instructions as indices, to achieve O(1) complexity.

I understand this might be beyond the scope of this project, but I still wanted to point this out because of the educational value.

typedef enum
{
    INST_NOP = 0,
    // ...
} Inst;

Err handle_nop(Bm *bm)
{
    bm->ip++;
    return ERR_OK;
}

// ...

Err (*inst_map[])(Bm *) = {
    handle_nop,
    // ...
};

Err bm_execute_inst(Bm *bm)
{
    // ...
    return inst_map[bm->inst.type](bm);
}

introduced in #76

introduced in #37

So you could write -DN=30 in the command line instead of %bind N 30 in the source.
(I don't know about -D. Maybe even without a flag, just N=30)

two Rs btw

For example

%bind secret "Uryyb, jbeyq! Sebz EBG13." ; the secret

raises error:

examples/rot13.basm:4: ERROR: `"Uryyb, jbeyq! Sebz EBG13." ; the secret` is not a number

introduced in #37

introduced in #59

https://piotrduperas.com/posts/nan-boxing/

Dear BMers

Imagine an extension of the Fibonacci implementation, where we would pass an argument N (numeric) "into" it.

How would we support it ?

One approach is to have access to argument count and vector, e.g. some form of global variable, like $argc and $argv.

Another way would be to be able to call call some OS input reading function.

A third one would be patching some operand between BM instantiation and its execution.

Say we had the code below, with the initial push instruction to be 10.
Changing this operand to 23 could be accomplished with:

./bme -i examples/fib.bm --patch-at 0 --patch-operand 23

or

./bme -i examples/fib.bm --patch 0:operand 23

Here's an incomplete Fibonacci implementation:

counter:
     push 10   # operand could be changed before execution

# First two Fibonacci numbers
   push 0
   push 1

loop:
   # check is counter is now 0
   # .......
   jmp_if halt
   dup 1
   dup 1
   plusi
   # decrement counter
   # ......
   jmp loop

halt:
   halt

introduce in #65

introduced in #49