Make stable_sort_primitive message less authoritative

[jongiddy]

Description

The current message for stable_sort_primitive says:

an unstable sort would perform faster without any observable difference for this data type
for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#stable_sort_primitive

Although it is a reasonable guideline and typically true for random data, it is not always true. For example, the append benchmark in https://gist.github.com/jongiddy/b29ce7d2251775dcf24c595bc245891b demonstrates that stable sort can be significantly faster for a scenario where a sorted vector is extended with new data and resorted.

running 6 tests
test almost_sorted_stable   ... bench:      12,290 ns/iter (+/- 97)
test almost_sorted_unstable ... bench:       3,392 ns/iter (+/- 74)
test append_stable          ... bench:       2,545 ns/iter (+/- 19)
test append_unstable        ... bench:       8,450 ns/iter (+/- 236)
test random_stable          ... bench:      16,803 ns/iter (+/- 396)
test random_unstable        ... bench:       6,149 ns/iter (+/- 156)

Because of this, the initial message should be less definite:

an unstable sort typically performs faster without any observable difference for this data type
for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#stable_sort_primitive

and the linked message should also be less definite, and suggest benchmarking specific situations.

Version

No response

Additional Labels

No response

[camsteffen]

Bummer, that also makes it a false positive and may warrant moving the lint to pedantic.

[spiral-ladder]

@rustbot claim