Repository navigation
Open
Conversation
A package that republishes data from a published article lists the article DOI in DESCRIPTION as X-schema.org-isBasedOn, next to the keywords and coverage fields. update_citation() looks the DOI up at doi.org and writes it as a CITATION.cff reference, with a message that asks users to cite both the data package and the article. inst/CITATION carries both entries under that message. The package stays the work that GitHub's "Cite this repository" shows. update_metadata() writes the same DOIs as schema.org isBasedOn. The field replaces the Config/washr/derived-from proposal in the issue, because the 2026-09-02 amendment of the canonical sources decision keeps schema.org facts under the X-schema.org- prefix. References that do not come from the field are dropped, so a rerun gives the same files. A DOI that cannot be looked up keeps its entry from the existing CITATION.cff. Closes #134 Assisted-by: Claude claude-fable-5-1
dev carries unreleased work for 1.2.0 (the source article citation, #134) on top of the 1.1.1 release, and still said 1.1.1. The development heading in NEWS.md is already in place. Assisted-by: Claude claude-fable-5-1
The washr skill (#115) parses the output of every function it runs, so messages need one structure and return values need to be real. Messages now go through three internal helpers on top of cli: a success line for a file written, an info line for something kept or skipped, an arrow line for the next step left to the user. Errors use cli::cli_abort() with a hint, and the three copies of the working directory check became check_pkg_root(), which names the function the user called. Every exported function returns what it wrote, invisibly, and @returns says so. options(washr.quiet = TRUE) silences washr's messages and the usethis helpers it calls, never warnings or errors. The usethis::ui_*() calls are gone (75 of them). usethis stays in Imports for the use_*() helpers. Closes #84 Assisted-by: Claude claude-fable-5-1
use_brand() wrote _brand.yml and the logo directory at the top level without adding them to .Rbuildignore, so the next R CMD check listed both as non-standard top-level files. The directory names come from the logo paths, in case the brand repository moves the files. Closes #133 Assisted-by: Claude claude-fable-5-1
washr is listed in the openwashdata R-universe registry, which serves a build of the development version that installs without a compiler toolchain. The README and the vignette now say what CRAN, R-universe and GitHub each give you. Closes #131 Assisted-by: Claude claude-fable-5-1
Zenodo's GitHub integration reads only a few fields of CITATION.cff and has no mapping for its type, so every data package release landed as Software in no community, and the record had to be edited by hand. When a repository carries a .zenodo.json, Zenodo uses that file for the record in place of CITATION.cff. update_zenodo_json() writes the file from the same DESCRIPTION the citation files come from: title, description, upload_type dataset, creators (aut and cre) with ORCID and affiliation, the license as Zenodo's identifier, keywords, version, the community, and the source articles of X-schema.org-isBasedOn as related identifiers with the relation isDerivedFrom. update_citation() calls it, so one step writes all three files and they cannot drift apart. The file is written as bytes (UTF-8, LF), so a second run is identical on every platform. The community is the first value read through washr_config(), the new reader for Config/washr/<key> fields in DESCRIPTION (#81). It defaults to openwashdata, takes several values, and "none" for no community. The license URLs of the JSON-LD and the Zenodo identifiers now share one table, license_map, so the two cannot disagree about which licenses washr knows. An unknown license is left out of the file, with a note that Zenodo then falls back to the license GitHub detects, or CC0 1.0 for a dataset. Closes #56 Assisted-by: Claude claude-fable-5-1
Until now a changed dataset left two choices: edit dictionary.csv by hand, or delete it and lose every description, because setup_dictionary() refuses an existing file and fill_dictionary() rebuilds the five columns from the data without reading what is there. update_dictionary() reads the file, merges it with the variables in data/ and writes only when the content differs. Rows are matched on file_name and variable_name. A matched row keeps its description and every other cell, including columns the user added (unit, allowed_values), which is the round trip openwashdata/pkgreview#37 waits for. Only variable_type is refreshed. New variables get an empty row, removed ones are named, and when a data object has both, the cells written for the removed variable are printed, since a rename cannot be told from a removal plus an addition. The file is read cell by cell as text, so codes and the string NA come back as typed, and it is written as UTF-8 bytes in the layout of write.csv(), independent of the session encoding. A file whose content is current is left untouched, whatever its quoting or line ends. The "already exists" error of setup_dictionary() now names the new function. Closes #13 Assisted-by: Claude claude-fable-5-1
The wiring rewrote _pkgdown.yml through the yaml package, which dropped the comments of the template and folded the funding text over two lines. The pkgreview check script looks for that text verbatim, so a freshly scaffolded package failed the funding item after use_brand(). The scaffold check found it on its first run. The brand lines are now inserted under the template block and the rest of the file stays as written. The yaml rewrite remains as the fallback when the template block already carries bslib settings or the edited file does not parse to the intended configuration. Refs #129 Assisted-by: Claude claude-fable-5-1
Nothing tested the claim that a freshly scaffolded package passes the required tier of the review standard, or that a second run of the steps changes no file. The new workflow builds a fixture with every washr step in vignette order, runs the re-runnable steps again and compares the tree, then runs pkgreview-check.R at the tag recorded in PKGREVIEW_FLOOR with the openwashdata profile. dev/release-checklist.md writes down the release steps from #78 and adds the floor bump. Closes #129 Assisted-by: Claude claude-fable-5-1
A skipped workflow step used to surface as an opaque error in a later function, and the guide had to warn about it. The new function reads the package and reports, item by item, what is in place, what is missing and which step closes each gap. It writes nothing. The items are the mechanical part of the openwashdata review standard: the metadata, docs and tests lines that pkgreview froze in its check script for this function (openwashdata/pkgreview#66), the dictionary checks pkgreview offered to hand over, and three items of washr's own (placeholder text in DESCRIPTION, data in data/, a help page per data object). Wording and details follow the script, so pkgreview can swap its frozen sections for one call and map the rows one to one. On the pkgreview fixture, its nine cases and three published packages, every mapped line agrees in status and detail. The result is a data frame of class washr_readiness with one row per item (id, area, check, status, detail, fix) and the attributes ready, washr_version and package. An item that cannot be decided is "na" with the reason, never a failure. The profile argument carries what an organisation expects of the site (analytics, Pages URL pattern, funding text, required keywords, brand), which pkgreview fills from its org profile. The report goes through cli and respects washr.quiet. Each item is one internal function, so a later step can run a single gate. Wiring those gates into the workflow functions is left for a follow-up. Closes #82 Assisted-by: Claude claude-fable-5-1
Zenodo takes the record of a release from .zenodo.json, version included, so a file that lags behind DESCRIPTION files the release under old values. The readiness report gains an item that compares the file with what DESCRIPTION gives now. The dictionary coverage item points at update_dictionary(). NEWS, the reference index and the README name the function, and the scaffold check stops when the report of the fixture has a failing item. Refs #82, #56, #13 Assisted-by: Claude claude-fable-5-1
The first scaffold-check run stopped at devtools::build_readme(): the README template loads dplyr, stringr, gt and kableExtra, which are not dependencies of washr. Refs #129 Assisted-by: Claude claude-fable-5-1
washr carried about thirty openwashdata facts in its functions and templates, so a group outside openwashdata could not publish with it without editing the output by hand. The organisation, the repository and the license are now read from the standard DESCRIPTION fields, and the values that have no standard field live in Config/washr fields that update_description() writes once and never overwrites: funding, analytics-domain, doi-provider, zenodo-community and brand-source, plus pages-domain for a site outside <org>.github.io. A field set to none switches its feature off. For a package under another organisation the funding, analytics, community and brand fields are written as none, because the openwashdata values would make false statements there. Config/washr/version records the washr version. The README and _pkgdown.yml templates, the DOI badge, .zenodo.json, use_brand() and the default profile of check_publication_readiness() read these values. For an openwashdata package the output is unchanged; the scaffold check still passes the review standard. use_brand() installs the brand at the latest release tag of the brand repository and records it in Config/washr/brand, where it copied from the main branch before. The tag comes from the redirect of the releases page, because the GitHub API rate limit is exhausted quickly on a shared network address. The vignette gains a section for groups outside openwashdata and a note on packages with more than one dataset. Closes #81 Closes #128 Refs #103 Assisted-by: Claude claude-fable-5-1
utils::write.csv() writes CRLF line ends on Windows whatever eol says, so the reference file of the byte comparison differed from the LF file that update_dictionary() writes on every platform. R CMD check failed on the Windows runner only. Refs #13 Assisted-by: Claude claude-fable-5-1
An independent review of the 1.2.0 candidate confirmed nine defects by running them. Each has a regression test in test_edge_cases.R. - update_description(github_user = ) after a first run left the package under the old organisation. The repository in URL is now replaced and the Config/washr fields that still hold the old defaults follow. - The usethis helpers wrote into the package of an earlier call when only the working directory had changed. Every exported function now sets the active project to the working directory first. - The funding item of the readiness report compared raw lines, so a quoted or folded funding text failed. It now also reads the value as YAML does. - The dictionary coverage item ignored the data object a variable belongs to. - A renamed data object lost its descriptions without the rename hint. - A repository URL with a .git suffix, a fragment, a trailing slash or www. gave a wrong bug report link or a second URL entry. - setup_readme(force = TRUE) deleted README.Rmd before it knew it could write a new one. - washr.quiet left the cffr messages of update_citation(). - update_description(file = "DESCRIPTION") skipped the license. Also from the agents that built on the candidate: a subset of the readiness report is a plain data frame (a column subset printed a wrong verdict), text files with invalid UTF-8 bytes are still matched, and a package of another organisation is held to its own license. Refs #81, #82, #13 Assisted-by: Claude claude-fable-5-1
Version and date, the NEWS section grouped the way 1.1.0 was, and the cran-comments section. The submission itself waits for the guide sync and is done by the maintainer. Assisted-by: Claude claude-fable-5-1
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
check_publication_readiness()(a read-only report of what a package still lacks, consumed by pkgreview),update_dictionary()(keeps descriptions and extra columns after a data change) andupdate_zenodo_json()(files a release on Zenodo as a Dataset in the community, called byupdate_citation()).options(washr.quiet = TRUE)silences the messages.URL, license fromLicense, andConfig/washr/fields for funding, analytics, DOI provider, Zenodo community and brand. Output for an openwashdata package is unchanged.use_brand()installs the brand at its latest release tag and records it in DESCRIPTION.scaffold-check, builds a fixture package with every step, runs the steps twice, and runs the pkgreview check script at the version inPKGREVIEW_FLOOR.Closes #84
Closes #82
Closes #81
Closes #128
Closes #13
Closes #56
Closes #129
Closes #131
Closes #133
Closes #134
Closes #122
Not in this PR: the washr skill (#115, #83) lives in the pkgreview plugin on its branch
feat/washr-1.2.0, the guide rework (#85, #80) waits for the new guide name, and the multi dataset loop of #103 is in the skill while the vignette carries the note.Test plan
devtools::test(): 640 pass, 0 faildevtools::check()with the CRAN settings: 0 errors, 0 warnings, 0 notesurlchecker::url_check(): all URLs correctpkgdown::check_pkgdown(): no problemsdevtools::submit_cran()frommainand the confirmation email (maintainer)v1.2.0,usethis::use_dev_version()ondev, syncdevwithmainPKGREVIEW_FLOOR