Remove bottlenecks for styler

[krlmlr]

Now running ~3x faster on the tidyr package (60s -> 20s), CC @lorenzwalthert.

[codecov]

Codecov Report

Merging #348 into master will increase coverage by 4.18%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #348      +/-   ##
==========================================
+ Coverage   87.43%   91.61%   +4.18%     
==========================================
  Files          24       23       -1     
  Lines        1122     1050      -72     
==========================================
- Hits          981      962      -19     
+ Misses        141       88      -53

Impacted Files	Coverage Δ
R/as_tibble.R	100% <100%> (ø)	⬆️
R/new.R	100% <100%> (ø)	⬆️
R/tibble.R	90.38% <100%> (-7.77%)	⬇️
R/tbl-df.r	98.48% <100%> (-1.52%)	⬇️
R/add.R	100% <100%> (ø)	⬆️
R/tribble.R	100% <100%> (ø)	⬆️
R/enframe.R	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update dd08491...bd6af96. Read the comment docs.

[lorenzwalthert]

Ok cool. Need to check how much performance styler gains through these optimizations.

[krlmlr]

@lionel-: Is there a helper for unclass(x))[[i]] in rlang?

[lionel-]

Not yet, maybe you can use .subset2()?

[krlmlr]

Of course, thanks!

[lorenzwalthert]

I just ran devtools::test() on styler with this patch and it takes only ~54% of the time compared to the current CRAN version of tibble! Pretty amazing.

[lionel-]

I'm thinking that the purrr compats should use purrr's C implementation to avoid the slowdown. It doesn't cost a lot to compile a little embedded C file.

[krlmlr]

Maybe move the elementary purrr functions to rlang, CC @hadley?

I've seen some slowdown due to rlang's very careful input checking. I'm really looking forward to the C implementation of the is_() family of functions ;-)

[lionel-]

sorry, it's planned but I haven't had time to do it yet :(

Expression capture and interpolation got a lot faster though, and I'll finish porting eval_tidy() to C for the next release as well.

[lionel-]

Most of rlang will be implemented in C eventually.

[krlmlr]

Thanks for the heads up, I really appreciate your work on rlang!

[lorenzwalthert]

Cool 😊

[lorenzwalthert]

Should we require a minimal version of tibble for styler to make sure the speed improvements take effect? Probably not, right?

[hadley]

I think it's reasonable to require it

[krlmlr]

I'm using update.packages(ask = FALSE); usethis::use_tidy_versions() to achieve this, see r-lib/usethis#179.