Deprecate signal_stage()

[DavisVaughan]

Closes #198 - Dark hackery no longer needed
Closes #199 - signalCondition() no longer needed
Closes #162 - No longer need this if signal_stage() is gone
Closes #91 - No longer need this if signal_stage() is gone
Closes #176 - It's very fast now
Closes #158 - No longer needed if we have no way to report on superseded functions.

This PR soft-deprecates signal_stage() and removes usage of it from deprecate_soft(), deprecate_warn(), and deprecate_stop(), so they are now very fast after all my other work here.

It also officially soft-deprecates signal_experimental() and signal_superseded(), so they can eventually be removed alongside signal_stage().

---

TLDR: In the tidyverse we should really encourage including a fixed id in our deprecate_soft() and deprecate_warn() calls. It makes them much much faster.

If you're using id and aren't using always = TRUE, it is VERY fast now. This is what dplyr and friends can start using everywhere for all deprecate_soft() and deprecate_warn() calls.

cross::bench_versions(
  pkgs = c(
    "lifecycle",
    "r-lib/lifecycle",
    "r-lib/lifecycle#202"
  ),
  \() {
    library(lifecycle)

    # trigger the per session warning once
    deprecate_soft("1.1.0", I("my thing"), details = "because", id = "foo")

    bench::mark(
      deprecate_soft("1.1.0", I("my thing"), details = "because", id = "foo"),
      iterations = 100000
    )
  }
)
#> # A tibble: 3 × 7
#>   pkg                 expression                               min   median `itr/sec` mem_alloc `gc/sec`
#>   <chr>               <bch:expr>                          <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 lifecycle           "deprecate_soft(\"1.1.0\", I(\"my … 145.22µs 154.53µs     6339.    8.55KB     44.2
#> 2 r-lib/lifecycle     "deprecate_soft(\"1.1.0\", I(\"my …  27.27µs  28.76µs    33428.    8.56KB     70.3
#> 3 r-lib/lifecycle#202 "deprecate_soft(\"1.1.0\", I(\"my …   5.54µs   5.86µs   160663.    8.56KB    183.

If you aren't using id, then we have to recreate your message every time to have a unique key, so it's a bit slower, but still quite a bit faster than where we were!

Here's an example with AsIs messages that don't require any what expression parsing

cross::bench_versions(
  pkgs = c(
    "lifecycle",
    "r-lib/lifecycle",
    "r-lib/lifecycle#202"
  ),
  \() {
    library(lifecycle)

    # trigger the per session warning once
    deprecate_soft("1.1.0", I("my thing"), details = "because")

    bench::mark(
      deprecate_soft("1.1.0", I("my thing"), details = "because"),
      iterations = 100000
    )
  }
)
#> # A tibble: 3 × 7
#>   pkg                 expression                                min  median `itr/sec` mem_alloc `gc/sec`
#>   <chr>               <bch:expr>                            <bch:t> <bch:t>     <dbl> <bch:byt>    <dbl>
#> 1 lifecycle           "deprecate_soft(\"1.1.0\", I(\"my th… 225.1µs   237µs     4138.    8.55KB     43.0
#> 2 r-lib/lifecycle     "deprecate_soft(\"1.1.0\", I(\"my th…  38.6µs  40.1µs    24288.    8.56KB     61.8
#> 3 r-lib/lifecycle#202 "deprecate_soft(\"1.1.0\", I(\"my th…  17.5µs  18.4µs    52918.    8.56KB     86.9

If you are using the expression parsing features of lifecycle and you aren't using a static id, then it's still not super fast. But remember this problem goes away entirely if you use a static id!

cross::bench_versions(
  pkgs = c(
    "lifecycle",
    "r-lib/lifecycle",
    "r-lib/lifecycle#202"
  ),
  \() {
    library(lifecycle)

    # trigger the per session warning once
    deprecate_soft("1.1.0", "my_fn(arg =)", "my_fn()")

    bench::mark(
      deprecate_soft("1.1.0", "my_fn(arg =)", "my_fn()"),
      iterations = 100000
    )
  }
)
#> # A tibble: 3 × 7
#>   pkg                 expression                                min  median `itr/sec` mem_alloc `gc/sec`
#>   <chr>               <bch:expr>                            <bch:t> <bch:t>     <dbl> <bch:byt>    <dbl>
#> 1 lifecycle           "deprecate_soft(\"1.1.0\", \"my_fn(a… 344.8µs 363.5µs     2716.    8.55KB     41.8
#> 2 r-lib/lifecycle     "deprecate_soft(\"1.1.0\", \"my_fn(a…  92.2µs  96.3µs    10125.    8.56KB     46.5
#> 3 r-lib/lifecycle#202 "deprecate_soft(\"1.1.0\", \"my_fn(a…    54µs  55.8µs    17425.    8.56KB     53.1

[DavisVaughan]

This test was super broken and wasn't testing what we thought it was. Mostly due to the deprecate_warn() calls needing to be wrapped in an actual function to simulate real usage and make them warn the right way.

[hadley]

Make sense to me.

[lionel-]

If you don't supply a what but still supply an id it's still very fast right?

Relatedly I think it's worth adding an annotation or comment of some kind to deprecate_warn0() that msg is passed lazily. The default assumption for R code should be that the timing of evaluation doesn't matter, so going against that should be declared/documented.

[lionel-]

(I'm not sure it's worth deprecating these functions.)

[DavisVaughan]

Decided over Zed to go ahead and soft-deprecate them with the goal of having an easy removal in a few years