Give a friendlier failure when lacking knitr

[MichaelChirico]

Closes #1266

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if b7ad523 is merged into main:

• ✔️cache_applying: 28.5ms -> 29.4ms [-0.28%, +6.68%]
• ✔️cache_recording: 433ms -> 435ms [-0.67%, +1.38%]
• ✔️without_cache: 1.02s -> 1.02s [-1.86%, +0.35%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[lorenzwalthert]

LGTM, thanks.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 2f971a8 is merged into main:

• ❗🐌cache_applying: 26.9ms -> 38.4ms [+38.73%, +46.91%]
• ❗🐌cache_recording: 413ms -> 425ms [+1.42%, +4.55%]
• ❗🐌without_cache: 929ms -> 939ms [+0.51%, +1.75%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[lorenzwalthert]

Just weird that this seems to slow down things…

[lorenzwalthert]

I triggered touchstone again to see if the problem persist.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 2f971a8 is merged into main:

• ❗🐌cache_applying: 26.9ms -> 39.9ms [+45.76%, +50.52%]
• ❗🐌cache_recording: 403ms -> 414ms [+2.12%, +3.44%]
• ✔️without_cache: 1000ms -> 1s [-1.15%, +1.55%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[lorenzwalthert]

Hmm... I think the speed test might actually be correct. Can we do the test for {knitr} more high level once, instead of every time we call get_knitr_pattern(), which seems quite a lot?

[lorenzwalthert]

Sorry I just realised we should not merge this PR as is.

[lorenzwalthert]

Also, @MichaelChirico, it would be great to fix this in the next 5 days, as I am planning to submit a new version of {styler} to CRAN.

[MichaelChirico]

I'm not too familiar with package internals so I spent only a few minutes trying to pin down where is the right place to move the check, I guess this is good because it avoids the iterated calls here:

styler/R/transform-code.R

Line 121 in fe6ae64

purrr::map2(starts, ends, finalize_raw_chunks,

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 377d573 is merged into main:

• ❗🐌cache_applying: 26.9ms -> 28.2ms [+2.65%, +6.73%]
• ✔️cache_recording: 403ms -> 405ms [-0.43%, +1.36%]
• ✔️without_cache: 949ms -> 946ms [-0.81%, +0.33%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[lorenzwalthert]

Much better, thanks @MichaelChirico 🥳. +2% for the cache_applying we can tolerate I think, since it's ultra fast anyways.