Numerical precision of log10.integer64() is not very satisfactory

[MichaelChirico]

expect_identical(log10(as.integer64(c(1L, 10L, 100L, 1000L))), as.numeric(0:3))
# Error: log10(as.integer64(c(1L, 10L, 100L, 1000L))) (`actual`) not identical to as.numeric(0:3) (`expected`).
#
#     actual             | expected              
# [1] 0.0000000000000000 | 0.0000000000000000 [1]
# [2] 1.0000000000000000 | 1.0000000000000000 [2]
# [3] 2.0000000000000000 | 2.0000000000000000 [3]
# [4] 2.9999999999999996 - 3.0000000000000000 [4]

c.f. the equivalent for integer/numeric input:

expect_identical(log10(c(1, 10, 100, 1000)), as.numeric(0:3))    # pass
expect_identical(log10(c(1L, 10L, 100L, 1000L)), as.numeric(0:3) # pass

[MichaelChirico]

The core of the implementation is:

bit64/src/integer64.h

Line 255 in c558eab

ret =(double) log10l((long double)e1); \

So basically, it's between the (long double) cast of long long and the (double) cast of the long double output of log10l().

Don't think it's worth tinkering with it further.