allow geom_sf to accept line parameters (#2826)

[alistaire47]

This closes #2826, allowing lineend, linejoin, and linemitre arguments to be passed through geom_sf. I copied the defaults directly from geom_path, so this shouldn't alter the results of current code.

The only change I had to make beyond what Claus mentioned on the issue was assigning the parameters to coord so they get passed through to sf_grob. There may be a simpler way to structure that. If so, let me know and I'll refactor.

For now, I decided against adding them as explicit parameters to geom_sf, as while it would make sense for lines, it would sound like they'd apply to polygons, too, which they wouldn't, and thus may cause more confusion than it'd remove.

This builds and performs fine. I didn't add any tests, as the geom_sf tests don't currently test its line behavior, but can if you like. I do get an error when checking due to the economics_long dataset. I didn't touch it, but it gives me a C++ error when printed, even though str works fine:

> economics_long
Error: Not compatible with STRSXP: [type=list].

I suspect this means I've got a dependency issue with tibble or something, but I can't figure out what. I presume that if it has been working in CI builds, it will continue to do so.

Apologies for the extra commits in there; I realized too late that I didn't rebase fully. They shouldn't do anything at this point, so if there's some git-fu to clean them out, I'd be happy to remove them.

[clauswilke]

Looks good to me. I have only one comment: Since sf_grob() is an internal function used only at that one point in the code, the lineend, linejoin, and linemitre could also simply be handed through as arguments to that function. Not sure whether there's a strong argument for or against that approach versus yours.

The extraneous git commits will disappear when we squash-merge, so don't worry about them.

[clauswilke]

Sorry, one more comment: Could you add a news bullet to NEWS.md with a brief description and your github user name and issue number?

[alistaire47]

Oh, I think that's better; those coord$lineend <- lineend lines feel like un-R-like boilerplate code.

The arguments will still get passed to sf_grob for point/multipoints, but ignored due to the logic therein, so it seems fine. I also noticed that the parameters get respected for polygons, which was unexpected but nice:

library(ggplot2)

ggplot(sf::st_polygon(list(matrix(c(0, 1, 2, 0, 0, 2, 1, 0), ncol = 2)))) + 
    geom_sf(size = 3, linejoin = 'mitre', fill = 'pink')

ggplot(sf::st_polygon(list(matrix(c(0, 1, 2, 0, 0, 2, 1, 0), ncol = 2)))) + 
    geom_sf(size = 3, linejoin = 'bevel', fill = 'lightgreen')

[alistaire47]

Oops, forgot, sorry! Updated.

[clauswilke]

Apologies for being picky. Would you mind adding a comma after linejoin in the news bullet. For context, see here: #2669 (comment)

[alistaire47]

Hah, that's now the smallest commit I've ever made.

[clauswilke]

@hadley Looks good to me.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/