Add workflow for benchmark

[BjarkeHautop]

Related to the discussion on Discord.

This adds a workflow that essentially just runs the relevant code from the performance article on PR and main.

It prints a table to compare the output of bench::mark() and it fails if any of the benchmarked functions have a median time that is at least 20% slower (fairly arbitrary, but I observed a 10% difference at one point, so it needs to be more than that to avoid false positives).

[BjarkeHautop]

Not sure why it doesn't print a table with units as it does locally? Here is how the table looks locally

# A tibble: 6 × 6
  Benchmark   PR_median PR_total_time Main_median Main_total_time Slowdown_percent_median
  <chr>        <bch:tm>      <bch:tm>    <bch:tm>        <bch:tm>                   <dbl>
1 parents        64.1µs         479ms        62µs           473ms                  3.39  
2 children       64.4µs         475ms      62.7µs           471ms                  2.71  
3 ancestors        72µs         470ms      72.4µs           465ms                 -0.552 
4 descendants    69.8µs         476ms      70.5µs           469ms                 -0.993 
5 subgraph      114.1µs         486ms       114µs           437ms                  0.0877
6 d_seperated    96.6µs         481ms      95.3µs           475ms                  1.36  

While on GA it looks like this:

# A tibble: 6 × 6
  Benchmark   PR_median    PR_total_time Main_median  Main_total_time
  <chr>       <bench_tm>   <bench_tm>    <bench_tm>   <bench_tm>     
1 parents     0.0001281491 0.4844711     0.0001298371 0.4840001      
2 children    0.0001274326 0.4841135     0.0001290006 0.4848658      
3 ancestors   0.0001475360 0.4741412     0.0001499145 0.4759974      
4 descendants 0.0001473451 0.4849910     0.0001503401 0.4813027      
5 subgraph    0.0002215331 0.4852312     0.0002308000 0.4834147      
6 d_seperated 0.0001925391 0.4805583     0.0001967761 0.4819401      
  Slowdown_percent_median
                    <dbl>
1                   -1.30
2                   -1.22
3                   -1.59
4                   -1.99
5                   -4.02
6                   -2.15

[jolars]

Nice idea! I am wondering... should we let the workflow comment on the PR too? That way you'd see the performance improvement/degradation directly?

[jolars]

LGTM, but I have a few small suggestions. No blockers though.

[jolars]

You should probably use the setup-r-dependencies action here, no?

[jolars]

Here, too, you should probably not call Rscript directly I think, or?

[jolars]

Maybe there's some nicer way to output a markdown table instead, what do you think?

[BjarkeHautop]

I am wondering... should we let the workflow comment on the PR too? That way you'd see the performance improvement/degradation directly?

Yeah, I thought about it, but then it should probably only do conditionally, I think (e.g, via a specific phrase in PR description/comment)? Otherwise, it probably gets annoying that it comes up when for a completely unrelated PR.

[jolars]

Yes, I think you're right!