I'm building a Tcl extension in c that needs to create new bignum Tcl_Objs (using Tcl_NewBignumObj), which takes an mp_int (libtommath's type). My extension (by default) calls Tcl's API through its stubs interface.
I include tclTomMath.h, and call Tcl_TomMath_InitStubs(interp, TCL_VERSION) in my extension's _Init. I'm not linking to an external libtommath (I want to use the version in the Tcl library that my extension will be loaded into). I don't define TCL_NO_TOMMATH_H, and include tcl.h before tclTomMath.h.
I haven't been able to find a single trace of any code that actually successfully does this. I read all the pages on the Tcl wiki that appeared to be relevant but left with more confusion than I started with.
My question is: what is the correct, idiomatic way for a stubs-enabled extension to create the mp_int argument that is passed to Tcl_NewBigNumObj? I also need to perform some minimal arithmetic on the mp_int before handing it to Tcl_NewBigNumObj, outside the range of Tcl_WideInt, so relying on the mp_int defined by tclTomMath.h if TCL_NO_TOMMATH_H is defined is not enough.
My Tcl sources (recent 8.7) do not install tommath.h by make install (Linux). If I don't copy it manually to the install headers location, compilation fails because tclTomMath.h tries to include it. If I do copy it, compilation succeeds, but loading my extension fails with missing symbols like undefined symbol: mp_error_to_string. mp_error_to_string does not appear to be in the subset of the tommath API that the Tcl fork's stub interface exposes. Is reporting generic error messages best the caller can do in this situation?
Sneaky bonus question: what would be the portable, idiomatic way to load the value -1-(2**64-1) (thanks CBOR...) into that mp_int?