Re: [Libjit] Creating constant values of different types

[Aleksey Demakov]

nint stands for "native int". This is similar to intptr_t in C but
IIRC was inspired by CLI/CLR (as libjit was created within dotGNU
project).

On Mon, Aug 20, 2018 at 10:11 PM Eli Zaretskii <address@hidden> wrote:
>
> There seem to be some hidden assumptions/dependencies between the
> various jit_value_create_TYPE_constant functions and the TYPE argument
> they accept.
>
> For example, jit_value_create_long_constant is documented to always
> create a 64-bit integer value, but it actually stores with the value
> the TYPE passed to it as its last argument, so one could put any type
> there.  If you do use type other than jit_long or jit_ulong, then
> getting the value back, e.g. with jit_value_get_long_constant will
> fail, and using other functions, like jit_value_get_nint_constant,
> will work on 64-bit platforms, but silently produce the wrong value on
> 32-bit ones.
>
> As an example,
>
>   jit_value_t v1 = jit_value_create_long_constant (f, x1, jit_type_nint);
>
> will create a value v1 whose type is jit_type_nint, so one might be
> tempted to use jit_value_get_nint_constant to obtain the value of x1
> back, which won't work.  Thus, it could be argued that
> jit_value_create_long_constant should not allow any TYPE except
> jit_long or jit_ulong.
>
> Another subtlety is the jit_nint type and jit_value_get_nint_constant
> that creates values of that type.  The documentation says that this
> function can be used to create "all pointer types" -- does that mean
> this is a 64-bit type on 64-bit hosts?  (If so, the "int" part of
> "nint" is mildly misleading, IMO.)
>
> Finally, the documentation mentions the sys_longlong type, but doesn't
> say what function can create constants of that type; I understand that
> only jit_value_create_long_constant can do that safely, is that true?
>
> Thanks.
>