fix symbolic/pynac.pyx doctests

[rwst]

Several doctests in symbolic/pynac.pyx iterate over range(get_ginac_serial(), get_ginac_serial()+100). You guess it (sigh), 100 is arbitrary and, right now with some new tickets that introduce new functions, it has become too small, leading to unrelated doctest fails in symbolic/pynac.pyx. This ticket should make sure that this does not happen again.

CC: @kcrisman

Component: symbolics

Author: Ralf Stephan

Branch/Commit: 52ccd47

Reviewer: Karl-Dieter Crisman

Issue created by migration from https://trac.sagemath.org/ticket/18257

[kcrisman]

comment:1

I agree that this is worth fixing permanently, but unless there is an obvious fix we shouldn't let this hold up adding more symbolic functions.

[rwst]

comment:2

Replying to @kcrisman:

I agree that this is worth fixing permanently, but unless there is an obvious fix we shouldn't let this hold up adding more symbolic functions.

This is not really difficult.

The central database is a dictionary globally defined in symbolic/function.pyx named sfunction_serial_dict. It has pairs with key of type unsigned and values of Function objects (which can be GinacFunctions or BuiltinFunctions). Accordingly this database is filled with all known Pynac functions (which have the GinacFunction façade) first and then those defined via BuiltinFunction. The unsigned key is the serial number which, on each new function registration, is incremented by one, namely in GiNaC::function::register_new()(https://github.com/pynac/pynac/blob/master/ginac/function.cpp#L1445) by just adding to the Pynac registry and reporting its size minus one. After the GiNaC functions have been added, the global GINAC_FN_SERIAL is set which is then accessed via get_ginac_serial() to get the start key of the Sage defined functions.

There is a problem if ever a function is deregistered. Can this happen? No mechanism in Pynac exists for this at the moment, so it's not a practical consideration.

So the solution is easy: instead of simply adding 100 to the start point, we try all keys from the start key to the latest key used---this need not be a static variable set on registration. Since functions are never deregistered the latest is, again, just the size of the GiNaC registry minus one. This can be had from a call to g_registered_functions().size() in symbolic/pynac.pyx.

cdef get_fn_serial_c():
    return g_registered_functions().size()

def get_fn_serial():
    return get_fn_serial_c()

sage: from sage.symbolic.pynac import get_fn_serial
sage: get_fn_serial()
127
sage: from sage.symbolic.pynac import get_ginac_serial
sage: get_ginac_serial()
40
sage: from sage.symbolic.function import get_sfunction_from_serial
sage: for i in range(get_ginac_serial(), get_fn_serial()):
    print get_sfunction_from_serial(i)
... // prints all known functions
sage: print get_sfunction_from_serial(get_fn_serial()+1)
None

[rwst]

comment:3

As they say the first thing that goes down the drain is the planning. The above solution does not account for user defined Python functions that do not inherit from BuiltinFunction or Function.

[rwst]

Branch: u/rws/fix_symbolic_pynac_pyx_doctests

[rwst]

Commit: 13ba99d

[rwst]

comment:5

No, that was an off-by-one error. Please review.

---

New commits:

13ba99d

18257: fix symbolic/pynac.pyx doctests