new `use_tidy_style()` #197

lorenzwalthert · 2018-01-08T18:37:51Z

Closes #72. I am not very familiar with the usethis source code, so let me know if there were some conventions I did not follow. Also, I am not sure if the warning (which I just copied from styler) in the documentation of the function is needed.
cc: @krlmlr.

jennybc · 2018-01-08T19:13:48Z

Just took a quick look. One thing I already know it needs is a check re: clean git status. See the source for use_dev_version() for some inspiration.

lorenzwalthert · 2018-01-08T19:16:15Z

thanks, I will do so.

jennybc · 2018-01-08T19:22:22Z

Can you also take a look at the AppVeyor failure?

lorenzwalthert · 2018-01-08T19:26:06Z

Yes, I try to understand what's going on because locally, with devtools::test() I get no errors. Some path issues I suppose. The problem is that I also don't fully understand scoped_temporary_package().
As far as the styling goes: In #72, it was suggested to basically alias styler::style_pkg(). However, usethis can also be used for non-package projects, right? So we should probably dispatch internally between style_pkg() and style_dir().

jennybc · 2018-01-08T19:29:09Z

However, usethis can also be used for non-package projects, right? So we should probably dispatch internally between style_pkg() and style_dir().

Yep, agree.

Yes, I try to understand what's going on because locally, with devtools::test() I get no errors. Some path issues I suppose.

It doesn't smell like a path problem FWIW. Also we're getting same failure on Travis and AppVeyor now. I think it's legit.

lorenzwalthert · 2018-01-08T19:29:26Z

For the uncommitted changes- test: I wondered whether I should create a function check_uncommitted_changes() instead of just c/p the code. Should I define it under R/helpers.R?

jennybc · 2018-01-08T19:32:12Z

For the uncommitted changes- test: I wondered whether I should create a function check_uncommitted_changes() instead of just c/p the code. Should I define it under R/helpers.R?

That's a good idea. I suggest you define it in R/git-utils.R.

lorenzwalthert · 2018-01-08T20:05:20Z

Ok, I took a look at use_dev_version() and tried to get an idea on how things should look. However, I still don't understand why the CI builds fail. I even get perfect R CMD CHECK and perfect rcmdcheck::rcmdcheck() locally and changed a few things but it still fails on travis and Appveyor.

lorenzwalthert · 2018-01-08T20:14:41Z

Also re-installed relevant dependencies from CRAN to make sure they match the CI configuration.

lorenzwalthert · 2018-01-08T20:25:25Z

Once the tests work I am happy to test case for a project that is not a package.

jennybc · 2018-01-08T20:35:14Z

It's not clear to me what the problem is. Suggest you remove the capture_output()s in the test to see if we learn anything useful from seeing a bit more output. You might also use simpler if()...else() logic instead of the do.call(). Will comment on the code itself ...

jennybc · 2018-01-08T20:35:32Z

NAMESPACE

 export(use_tidy_support)
 export(use_tidy_versions)
 export(use_travis)
 export(use_usethis)
 export(use_vignette)
+importFrom(styler,style_pkg)
+importFrom(styler,tidyverse_style)


We generally don't do this.

So should I just use :: whenever I need an external function?

jennybc · 2018-01-08T20:36:21Z

R/styler.R

+  check_is_package("use_tidy_style()")
+  check_uncommitted_changes()
+  styled <- do.call(
+    ifelse(is_package(), "style_pkg", "style_dir"),


I wouldn't use ifelse() this way.

In fact, I'd unwrap this into a more conventional if() ... else() construct.

lorenzwalthert · 2018-01-08T21:43:38Z

Sorry I did not realize earlier but maybe we should move use_tidy_style() to the other tidy_* functions in R/tidyverse.R and do the same with tests.

lorenzwalthert · 2018-01-12T23:27:31Z

For some reason the styling throws an error. Unfortunately, the transformation of the code within style_pkg() is done with enc::transform_lines_enc() so we can't easily print the source of the problem by changing the source code of usethis. Any change anyone could try to run the tests locally and see if they get the same problem too?

jennybc · 2018-01-15T20:57:00Z

tests/testthat/helper.R

-    package = capture_output(create_package(dir, rstudio = rstudio, open = FALSE)),
-    project = capture_output(create_project(dir, rstudio = rstudio, open = FALSE))
+    package = testthat::capture_output(create_package(dir, rstudio = rstudio, open = FALSE)),
+    project = testthat::capture_output(create_project(dir, rstudio = rstudio, open = FALSE))


These aren't necessary and aren't used elsewhere in the tests. Can you remove?

In terms of test development, I load testthat in all interactive sessions in .Rprofile FWIW.

Yes they are not necessary if testthat is loaded. I thought it might be good practice to not assume this is the case so I added them. But I will remove them in that case. As it stands if now, the PR is anyways not intended to be merged, the primary goal was to make the unit tests passing, which was not possible.

ok, removed.

jennybc · 2018-01-15T21:33:39Z

I'm about to run the tests locally on Windows and see if I learn anything.

lorenzwalthert · 2018-01-15T21:37:37Z

Thanks. Also, once we get them running, we can squash all the commits we needed to get there to keep a clean history I suggest.

jennybc · 2018-01-15T22:07:10Z

I can replicate but not explain the failure on Windows. But only via devtools::check(). Tests pass with devtools::test() and via R CMD build + R CMD check. It's a peculiar combination of results.

If I replace use_tidy_style() in the test with write_utf8(path_to_bad_style, "a <- 2\n"), tests pass. So this is a true result, in some sense. I think the file really is not getting restyled but I don't know why.

jennybc · 2018-01-15T22:13:33Z

You might need to make a debugging branch of styler that emits more info from style_pkg() and maybe prettify_pkg() and use the Remotes field in DESCRIPTION to install styler from that branch.

Alternatively, you could try to avoid the use of scoped_temporary_package() in your test, i.e. create the bare minimum file structure to convince these functions to work. Like, just fake making a package. Part of me wonders if this has something to do with withr, which is used in scoped_temporary_package() and in style_pkg().

Update: re avoiding scoped_temporary_package(). You could probably just inline a lot of that logic into your test, in this experiment. Let's try to eliminate withr as our problem.

lorenzwalthert · 2018-01-15T22:32:13Z

Ok. Thanks for helping me out with that. As you said, it seems really strange. I hope I can find time soon to look at it more in depth and follow some of the suggestions you outlined above to find the root of the problem. It would be nice if we could continue to use some of the existing infrastructure eventually though. I did also not think of withr as a potential cause, thanks for that hint.

jennybc · 2018-01-15T22:45:28Z

If it is a bad interaction between these two, nested uses of withr, that would be interesting to know. At this point, I suspect use_tidy_style() works just fine and we're now learning something about our tooling for package development and testing. But we can't merge this until we can test it. By the way, this should test the "non-package project" branch too, before we're done.

jimhester · 2018-01-17T20:16:00Z

The failures are because you have an implicit dependency on dplyr since you are using map_dfr()in styler

And purrr has a dependency on dplyr for map_dfr().

IMHO you need to remove the map_dfr() calls in styler (and remove dplyr from Suggests: in styler, so that these errors surface sooner)

lorenzwalthert · 2018-01-17T20:45:08Z

Thanks so much @jimhester, I would never have considered that. I assumed we have removed the dplyr dependency completely with r-lib/styler#182 but it was apparently only after that (as you linked above) where I introduced it again without taking notice. I'll try to fix it.

lorenzwalthert · 2018-01-17T20:58:09Z

@jennybc might that also affect the upcoming reprex release with styler in Suggests: because the dplyr dependency is missing? Also, it seems quite a coincidence that we found out about this problem because only if the expression to style contains an EQ_ASSIGN token (like in a = 2 that I implemented in a test), purrr::map_dfr() is used.

lorenzwalthert · 2018-01-17T21:15:07Z

I opened r-lib/styler#323 to discuss the implicit dplyr dependency and implications in general.

jennybc · 2018-01-17T21:29:11Z

might that also affect the upcoming reprex release with styler in Suggests

I now regard this as completely a styler problem 😁

Realistically, the vast vast majority of usethis and reprex users will have dplyr. So I won't let this block a CRAN update to usethis or reprex. Nothing has gotten worse, now we just know about a pre-existing problem.

But it would be good to straighten things out over in styler! And get rid of the dplyr dependency.

I propose that you change the test here to something that doesn't exercise the dplyr-calling code in styler, so we can move forward. If you're really fast, it can happen before we submit usethis -- TODAY.

jennybc · 2018-01-17T21:32:26Z

Any test that will make git commits needs to have this in it:

usethis/tests/testthat/test-use-readme.R

Lines 12 to 14 in da1a8c2

    
           # git2r::git2r::discover_repository() not working on R 3.1 (Travis) 
        
           skip_if(getRversion() < 3.2) 
        
           skip_if_no_git_config()

remove capture_output for now as suggested by reviewer to see what happens.

because git2r::discover_repository() not working on R 3.1.

lorenzwalthert · 2018-01-17T22:13:26Z

Ok, I think that's it. In particular, I reverted the testthat::capture_output() mask I introduced and added a test for a project.
Let me know what you think. I know there are many commits now so can we just squash and merge at the end? Otherwise I can do interactive rebase and try to squash individual commits together but if that's not required from your side will spare this effort 😊.

jennybc · 2018-01-17T22:17:10Z

This will be a squash and merge regardless. Let me just look it all over again.

lorenzwalthert · 2018-01-17T22:17:45Z

Ok, sure.

jennybc · 2018-01-17T22:19:23Z

R/dev-version.R

-      call. = FALSE
-    )
-  }
+  check_uncommitted_changes(proj_get())


Pretty sure this can just be check_uncommitted_changes(), right?

Yes, I was not sure because it appeared to me that in some places the path was set while in other places it was not, so I thought I go the safe way and add it. Will remove it then.

jennybc · 2018-01-17T22:20:42Z

R/styler.R

+    )
+  }
+  cat_line()
+  done("Styled package according to the tidyverse style guide")


Propose: "Project has been styled according to ..." to be less package specific.

jennybc · 2018-01-17T22:22:13Z

Green checks on travis!

Just change those two things.

I'm editing NEWS right now so will anticipate this for you.

lorenzwalthert · 2018-01-17T22:27:07Z

Done. On a related note, I think there are quite help files that are not up-to-date with the package source code at that stage but I have not committed anything because it's not related to this PR.

jennybc · 2018-01-17T22:31:09Z

I think there are quite help files that are not up-to-date with the package source code

I've been documenting a lot, so if you have concrete examples, please pass along. I'm not aware of a problem.

jennybc · 2018-01-17T22:31:13Z

Thanks!

lorenzwalthert · 2018-01-17T22:35:38Z

Thanks 😊. Ok, I think the reason for that was simply different roxygen versions (I don't use the dev version) and my changes had made things worse, so I think it's all good.

lorenzwalthert force-pushed the styler branch from 81daf59 to 840e53a Compare January 8, 2018 19:01

jennybc reviewed Jan 8, 2018

View reviewed changes

jennybc reviewed Jan 15, 2018

View reviewed changes

lorenzwalthert mentioned this pull request Jan 17, 2018

styler has an implicit dplyr dependency r-lib/styler#323

Closed

lorenzwalthert added 10 commits January 17, 2018 22:53

adapt use_dev_version() style

d819d7c

avoid EOF warning

d477e7e

switch to usethis:::proj_path()

4025533

adapt to review comments

a47318a

fiddle with tests...

f4b7f35

remove capture_output for now as suggested by reviewer to see what happens.

mask testthat call

ff445aa

printing

e31c255

fix test to not include dplyr-dependency edge case in styler.

b3ff15b

skip tests on R < 3.2

1f183f5

because git2r::discover_repository() not working on R 3.1.

remove print statements

c94dbb1

lorenzwalthert force-pushed the styler branch from d09db1c to c94dbb1 Compare January 17, 2018 21:53

lorenzwalthert added 3 commits January 17, 2018 22:56

fix bug for non-packages

75bf4ad

capture output

8407f1a

remove testthat mask

90695cb

lorenzwalthert force-pushed the styler branch from 28fa060 to cb186db Compare January 17, 2018 22:04

adapt help file and test description for non-packages

9b7230a

lorenzwalthert force-pushed the styler branch from cb186db to 9b7230a Compare January 17, 2018 22:06

jennybc reviewed Jan 17, 2018

View reviewed changes

final tweaks

aeca15b

jennybc merged commit 2abb042 into r-lib:master Jan 17, 2018

lorenzwalthert deleted the styler branch January 17, 2018 22:36

new use_tidy_style() #197

new use_tidy_style() #197

Conversation

lorenzwalthert commented Jan 8, 2018

jennybc commented Jan 8, 2018

lorenzwalthert commented Jan 8, 2018

jennybc commented Jan 8, 2018

lorenzwalthert commented Jan 8, 2018

jennybc commented Jan 8, 2018

lorenzwalthert commented Jan 8, 2018

jennybc commented Jan 8, 2018

lorenzwalthert commented Jan 8, 2018 • edited Loading

lorenzwalthert commented Jan 8, 2018

lorenzwalthert commented Jan 8, 2018

jennybc commented Jan 8, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lorenzwalthert commented Jan 8, 2018

lorenzwalthert commented Jan 12, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jennybc commented Jan 15, 2018

lorenzwalthert commented Jan 15, 2018

jennybc commented Jan 15, 2018 • edited Loading

jennybc commented Jan 15, 2018 • edited Loading

lorenzwalthert commented Jan 15, 2018

jennybc commented Jan 15, 2018

jimhester commented Jan 17, 2018 • edited Loading

lorenzwalthert commented Jan 17, 2018

lorenzwalthert commented Jan 17, 2018 • edited Loading

lorenzwalthert commented Jan 17, 2018

jennybc commented Jan 17, 2018 • edited Loading

jennybc commented Jan 17, 2018

lorenzwalthert commented Jan 17, 2018

jennybc commented Jan 17, 2018

lorenzwalthert commented Jan 17, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jennybc commented Jan 17, 2018

lorenzwalthert commented Jan 17, 2018

jennybc commented Jan 17, 2018

jennybc commented Jan 17, 2018

lorenzwalthert commented Jan 17, 2018

new `use_tidy_style()` #197

new `use_tidy_style()` #197

lorenzwalthert commented Jan 8, 2018 •

edited

Loading

lorenzwalthert commented Jan 12, 2018 •

edited

Loading

jennybc commented Jan 15, 2018 •

edited

Loading

jennybc commented Jan 15, 2018 •

edited

Loading

jimhester commented Jan 17, 2018 •

edited

Loading

lorenzwalthert commented Jan 17, 2018 •

edited

Loading

jennybc commented Jan 17, 2018 •

edited

Loading