This is how benchmark results would change (along with a 95% confidence interval in relative change) if c7aa214 is merged into main:

• ✔️default: 34.7s -> 32.6s [-16.5%, +4.71%]
• ✔️no_delays: 39.4s -> 42.4s [-2.87%, +18.22%]
• ✔️random_walk: 11.6s -> 10.6s [-33.01%, +14.58%]
• ✔️stationary: 23.9s -> 18.6s [-56.38%, +12.16%]
• ✔️uncertain: 49.8s -> 52s [-11.3%, +20.22%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

This is how benchmark results would change (along with a 95% confidence interval in relative change) if acf754f is merged into main:

• ✔️default: 40.5s -> 31.8s [-53.43%, +10.38%]
• ✔️no_delays: 44.4s -> 38.2s [-31.73%, +3.51%]
• ✔️random_walk: 10.2s -> 9.69s [-11.97%, +2.72%]
• ✔️stationary: 18.5s -> 19s [-12.28%, +18.18%]
• ✔️uncertain: 49.4s -> 50.1s [-11.1%, +13.94%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

Hmm how odd this works locally for me...

This is how benchmark results would change (along with a 95% confidence interval in relative change) if bcf40f3 is merged into main:

• ✔️default: 32.4s -> 32.7s [-15.73%, +17.45%]
• ✔️no_delays: 54.6s -> 46.3s [-75.01%, +44.6%]
• ✔️random_walk: 10.2s -> 10.4s [-17.42%, +21.59%]
• ✔️stationary: 18.6s -> 19.8s [-13.45%, +26.56%]
• ✔️uncertain: 53.8s -> 56.2s [-14.4%, +23.12%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

Hmm how odd this works locally for me...

Yes, strange, seems to call base::truncate over the loaded stan function for some reason.

Maybe its the order things are loaded. I've tried to hack around it by renaming the stan function just after its exposed. Not ideal but good enough I think.

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 1adf432 is merged into main:

• ✔️default: 31.9s -> 1.49m [-219.01%, +578.85%]
• ✔️no_delays: 42.3s -> 42.3s [-13%, +12.82%]
• 🚀random_walk: 10.3s -> 9.47s [-16.24%, -0.12%]
• ✔️stationary: 20.8s -> 18.8s [-21.69%, +2.55%]
• ✔️uncertain: 54.9s -> 50.9s [-21.19%, +6.57%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

I'm at loss at what's going on here - can't reproduce this locally either (even e.g. running testthat::test_local() in a fresh session, which should be equivalent).

We could rename the function which would be annoying but perhaps the easiest option, or ditch the tests, or try to understand what exactly is going on here (e.g. print the return values of the rstan::expose_stan_functions call).

We could rename the function which would be annoying but perhaps the easiest option

So I think we want tests over anything else so voted here to change the name to avoid the collision. Not ideal at all though obviously

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 6e5dff3 is merged into main:

• ✔️default: 32.5s -> 37.9s [-2.9%, +36.47%]
• ✔️no_delays: 43.8s -> 41.6s [-19.58%, +9.54%]
• ✔️random_walk: 9.5s -> 10.9s [-15.64%, +44.98%]
• ✔️stationary: 20.4s -> 18.8s [-22.9%, +7.13%]
• ✔️uncertain: 56.6s -> 54.9s [-21.08%, +15.03%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

@jamesmbaazam as @sbfnk is on holidays if you get chance to give this a look that would be great. I think its mostly there. I am unhappy about the need to name change truncate but am also not sure its worth the effort of dealing with blocking the base::truncate issue.

Note this PR also starts adding docs as in #740 and improves stan unit test coverage (no way I know of getting some like code coverage for this).

Thanks for the fix, Sam. I'm taking a look now.