Closed Jake-Moss closed 1 week ago
- coerce a mpoly from one context to another, either with an inferred mapping or a provided one.
We should check what generic rings does before adding any sort of coercion like this. I didn't add coercion in gh-218 but we could easily add some explicit coercion there.
we could easily add some explicit coercion there.
There are two functions related to this gr_set_other
and gr_ctx_cmp_coercion
.
There are two functions related to this
gr_set_other
andgr_ctx_cmp_coercion
.
I've taken a quick look at these and their mpoly variants.
The gr_ctx_cmp_coercion
function says "Returns 1 if coercing elements into ctx1 is more meaningful, and returns -1 otherwise.", it seems to measure this by the ordering of the gr_which_structure
enum. So if ctx1->which_ring
is defined later than ctx2->which_ring
then it is "more meaningful".
The coerce_to_context
method in this PR only works between mpoly contexts of the same type. In that cause we'd fall all the way through in this function and return 1
.
// flint/src/gr/cmp_coercion.c
int gr_ctx_cmp_coercion(gr_ctx_t ctx1, gr_ctx_t ctx2)
{
if (ctx1->which_ring < ctx2->which_ring)
return -1;
if (ctx1->which_ring > ctx2->which_ring)
return 1;
if (ctx1->which_ring == GR_CTX_GR_POLY)
{
return gr_ctx_cmp_coercion(POLYNOMIAL_ELEM_CTX(ctx1), POLYNOMIAL_ELEM_CTX(ctx2));
}
if (ctx1->which_ring == GR_CTX_GR_MAT)
{
return gr_ctx_cmp_coercion(MATRIX_CTX(ctx1)->base_ring, MATRIX_CTX(ctx2)->base_ring);
}
return 1;
}
// flint/src/gr.h
typedef enum
{
GR_CTX_FMPZ, GR_CTX_FMPQ, GR_CTX_FMPZI,
GR_CTX_FMPZ_MOD, GR_CTX_NMOD, GR_CTX_NMOD8, GR_CTX_NMOD32, GR_CTX_MPN_MOD,
GR_CTX_FQ, GR_CTX_FQ_NMOD, GR_CTX_FQ_ZECH,
GR_CTX_NF,
GR_CTX_REAL_ALGEBRAIC_QQBAR, GR_CTX_COMPLEX_ALGEBRAIC_QQBAR,
GR_CTX_REAL_ALGEBRAIC_CA, GR_CTX_COMPLEX_ALGEBRAIC_CA,
GR_CTX_RR_CA, GR_CTX_CC_CA,
GR_CTX_COMPLEX_EXTENDED_CA,
GR_CTX_RR_ARB, GR_CTX_CC_ACB,
GR_CTX_REAL_FLOAT_ARF, GR_CTX_COMPLEX_FLOAT_ACF,
GR_CTX_NFLOAT, GR_CTX_NFLOAT_COMPLEX,
GR_CTX_FMPZ_POLY, GR_CTX_FMPQ_POLY, GR_CTX_GR_POLY,
GR_CTX_FMPZ_MPOLY, GR_CTX_GR_MPOLY,
GR_CTX_FMPZ_MPOLY_Q,
GR_CTX_GR_SERIES, GR_CTX_SERIES_MOD_GR_POLY,
GR_CTX_GR_MAT,
GR_CTX_GR_VEC,
GR_CTX_PSL2Z, GR_CTX_DIRICHLET_GROUP, GR_CTX_PERM,
GR_CTX_FEXPR,
GR_CTX_UNKNOWN_DOMAIN,
GR_CTX_WHICH_STRUCTURE_TAB_SIZE
}
gr_which_structure;
Now looking at the gr_set_other
method, the fmpz_mpoly
gr defines this to be
// flint/src/gr/fmpz_mpoly.c
gr_method_tab_input _gr_fmpz_mpoly_methods_input[] =
{
// ...
{GR_METHOD_SET_OTHER, (gr_funcptr) _gr_fmpz_mpoly_set_other},
// ...
};
which directly sets the polynomial if the two contexts match in nvars and ordering
int
_gr_fmpz_mpoly_set_other(fmpz_mpoly_t res, gr_srcptr x, gr_ctx_t x_ctx, gr_ctx_t ctx)
{
if (x_ctx->which_ring == GR_CTX_FMPZ_MPOLY)
{
/* fmpz_mpoly_set_fmpz_poly */
if (MPOLYNOMIAL_MCTX(ctx)->minfo->nvars == MPOLYNOMIAL_MCTX(x_ctx)->minfo->nvars &&
MPOLYNOMIAL_MCTX(ctx)->minfo->ord == MPOLYNOMIAL_MCTX(x_ctx)->minfo->ord)
{
fmpz_mpoly_set(res, x, MPOLYNOMIAL_MCTX(ctx));
return GR_SUCCESS;
}
}
return gr_generic_set_other(res, x, x_ctx, ctx);
}
In any other case it falls back to
// flint/src/gr_generic/generic.c
int gr_generic_set_other(gr_ptr res, gr_srcptr x, gr_ctx_t xctx, gr_ctx_t ctx)
{
if (xctx == ctx)
{
return gr_set(res, x, ctx);
}
else if (xctx->which_ring == GR_CTX_FMPZ)
{
return gr_set_fmpz(res, x, ctx);
}
else if (xctx->which_ring == GR_CTX_FMPQ)
{
return gr_set_fmpq(res, x, ctx);
}
else if (xctx->which_ring == GR_CTX_FEXPR)
{
gr_vec_t vec;
fexpr_vec_t fvec;
int status;
fexpr_vec_init(fvec, 0);
gr_vec_init(vec, 0, ctx);
status = gr_set_fexpr(res, fvec, vec, x, ctx);
gr_vec_clear(vec, ctx);
fexpr_vec_clear(fvec);
return status;
}
else
{
return GR_UNABLE;
}
}
which attempts some easy settings and what seems to be a conversion to a symbolic expression, otherwise it returns GR_UNABLE
meaning the conversion is unimplemented.
GR_CTX_GR_MPOLY
doesn't define an explicit GR_METHOD_SET_OTHER
so I believe it falls back to the same generic method above.
It seems like the generic rings are only able to change mpoly context directly when the contexts match exactly. Maybe theres some method to go to a symbolic expression then back to another context, I'm not sure.
The coerce_to_context
here is able to convert a mpoly between contexts of the same python-flint type, i.e. fmpz_mpoly_ctx
to fmpz_mpoly_ctx
via the compose generator functions. It's able to handle a change in nvars by mapping any absent ones to 0, and a change in ordering as the *_mpoly_compose_mat
method used internally performs a sort, then combines like terms to normalise the polynomial before returning.
Sorry, I missed that this was marked as ready for review.
Sorry, I missed that this was marked as ready for review.
No worries at all
Okay, let's get this in.
This adds methods to
As well as some doc string updates and support for negative indecies in the
variable_to_index
method.Missing ATM:
fmpz_mod_mpoly_compose_fmpz_mod_mpoly_gen
function in Flint, PR here https://github.com/flintlib/flint/pull/2068 I've disabled support for it ATM and prevented the linker from complaining on import with a macro.