Inaccurate log10

[librasteve]

Expected Behaviour

log10(1000); #3 (Rat)

Actual Behaviour

log10(1000); #2.9999999999999996 (Num)

Rationale

raku is proud to handle decimals (and other rationals) in a way that avoids irrational results (geddit?)

python handles this "right"
https://gist.github.com/librasteve/fc2999c28654d219876c537abf554524

so does C
https://www.netlib.org/fdlibm/e_log10.c

Change Needed

looks like this:

MoarVM/src/core/interp.c

Line 967 in 75fe055

log(x);

could maybe add new OP something like this:

            OP(log10_n): {
                MVMnum64 x = GET_REG(cur_op, 2).n64;
                GET_REG(cur_op, 0).n64 =
#ifdef MVM_HAS_SUBSTANDARD_LOG
                    x < 0 ? MVM_num_nan(tc) :
#endif
                    log10(x);
                cur_op += 4;
                goto NEXT;
            }

• that is MoarVM is anyway using C operators - so why not use the specific one intended?
• would need to back out and repoint the higher up bit of code that converts a log10 raku op to a log MoarVM OP
• maybe there are other Rat oriented raku math operators to consider - ie. log2

[vushu]

@librasteve have you tried log(1000, 10) ?

[librasteve]

[3] > log(1000,10) #2.9999999999999996

same problem