Kirill Müller

Current state of affairs: https://rpubs.com/krlmlr/styler-prof-1 . This requires a cleanup of `Rprof.out` and a tweak to `visit_one()`, will share soon. Surprisingly, I didn't have to convert pipes (magrittr -> native)...

With #980 and #982 we require only a cleanup of `Rprof.out` to get https://rpubs.com/krlmlr/styler-prof-2: ``` r text Deactivated cache. Rprof(filter.callframes = TRUE, line.profiling = FALSE) for (i in 1:10) styler::style_text(text)...

The profiling with `%>%` replaced by `|>` might be a tad cleaner, I'm not sure: https://rpubs.com/krlmlr/styler-prof-3 After merging all relevant PRs, I believe a speed improvement by a factor of...

I'd only go for `ifelse()` if benchmarks suggest that this is useful. We want tibble subsetting to be fast, but it's a larger project. We can: - fix in styler...

A quick heads-up: I tested with https://github.com/tidyverse/tibble/pull/1469, which reverts to regular data.frame subsetting. - All styler tests still pass - For styler's `nest.R`, `style_text(scope = "none")` drops from 1.03s to...

A move to data.table seems like a paradigm shift -- from immutable data to one overarching mutable "parse data" structure. To me, this looks like a major refactoring, touching all...

- `furrr::future_map_lgl()` is a drop-in replacement for `map_lgl()`, gives a progress bar, and has seen an update very recently. It's a suggested package, the default is to revert to serial...

Thanks! Happy to review. The startup costs are a problem. On Linux we could start one child process with callr, which then could spawn subprocesses very efficiently with the multicore...

Not sure, why not add a line break after each full stop?

```r #' I try to avoid this problem altogether by adding a line #' break after each full stop. #' This means that edits in previous sentences do not #'...