Scope the empty-column filter to columns nobody chose

[ibuilder]

Reviewing the 2.0.5 diff turned up three defects, all in the column filter it introduced. Each was confirmed by rendering the shortcode, not by reading.

An explicit columns= list was filtered like any other

[procore_team columns="name,email_address"] rendered only Name. The column was named in the shortcode and removed with no header, no cell and no explanation. The filter's own docblock claimed the opposite — "A column an author asked for explicitly is always kept" — but nothing in the code distinguished an author's list from the registry defaults.

The filter now runs only over the default column set. A hand-written list is rendered verbatim, empty or not.

The filter could not do most of what it claimed

Format::cell() returns an em dash for a missing value in every format except email, so a test for an empty string could only ever match a suppressed email address — and exactly one column in the whole endpoint registry uses that format. Every other empty column survived as a full column of dashes, while the commit message ("a description or phone nobody has filled in"), CHANGELOG.md and readme.txt all described a feature that was not there.

The placeholder is now the named constant Format::PLACEHOLDER, and Format::is_blank() judges a cell by meaning rather than by length. [procore_rfis] against a project where no RFI has a due date now drops the column, as documented.

A column's fate depended on its neighbours

Because the never-render-an-empty-table fallback returned the unfiltered list, columns="email_address" kept the Email column while columns="name,email_address" dropped it. Same data, same setting, opposite outcome — adding a column to a working shortcode could silently remove a different one. Honouring explicit lists removes this at the root.

Tests

Three tests, one per defect. Each was run against the previous behaviour first and observed to fail before the fix was restored:

1) test_keeps_an_empty_column_the_author_asked_for
2) test_drops_a_default_column_no_record_populates
3) test_column_visibility_does_not_depend_on_other_columns
Tests: 106, Assertions: 399, Failures: 3.

Also

• The 2.0.5 changelog and readme entries are corrected where they claimed behaviour that never shipped.
• docs/shortcodes.md gains an Empty columns section stating what the filter does and where it stops.
• .distignore excludes /vendor rather than /vendor/bin. Every dependency is dev-only and the release workflow was passing an extra --exclude=vendor to compensate, so anyone building from .distignore alone shipped the whole dev tree.

Verification

• 106 tests, 401 assertions — pass
• PHPCS WordPress-Extra + WordPress-Docs — clean across 52 files
• Plugin Check against the staged distributable, all categories plus experimental checks at severity 1 — no errors, no warnings. Confirmed the checker was live by running it against the bundled Akismet and Hello Dolly, which it flags extensively.

🤖 Generated with Claude Code