Keep rows, cols and multiple off the wrapper div - #211

Closes #210.

rows, cols and multiple now render only on the input, not on the wrapper <div> as well.

Cause

Base::__initializeMeta() renames $key to "field" . $key, routes the value, then calls unset($this->__meta[$key]) — removing "fieldcols", which never existed. The original "cols" survives and __getMetaString() prints it on the wrapper.

Only the unprefixed form is hit; fieldcols skips the rename so its unset matches. Unprefixed is the intended form here — $__reservedfieldmeta exists to route these three keys, Element::setClass() reads $meta["multiple"] unprefixed to pick vf__one or vf__multiple, and examples/textarea.php documents "cols" / "rows".

Fix

Capture the key before the rename, unset that one.

That leaves __getMetaString()'s second guard with nothing to catch, and it was wrong anyway:

if (! in_array($key, array_merge($this->__reservedmeta, $this->__fieldmeta))) {

array_merge contributes __fieldmeta's values, not its keys, so a wrapper attribute vanished whenever an unrelated field meta value equalled its name — ['fieldplaceholder' => 'class'] erased the wrapper's own class="vf__optional". Merge dropped.

array_keys() is not a valid replacement: __fieldmeta always carries a class key, so it would suppress class on every wrapper.

Tests

955 tests, 1765 assertions, green. Three added, each confirmed to fail on master before the fix:

• TextareaTest::unprefixedRowsAndColsStayOffTheWrapper
• SelectTest::multipleMetaStaysOffTheWrapper
• ElementTest::wrapperAttributesSurviveAFieldMetaValueMatchingTheirName

Each asserts both halves — gone from the wrapper, still on the input — so the routing cannot be dropped to make them pass. ElementTest gained use HtmlAssertionsTrait; it had no DOM helper.

For reviewers

The __initializeMeta() docblock above the changed loop said $meta["labelstyle"] becomes $__fieldmeta["style"]. It becomes $__labelmeta["style"]. Corrected in place — say so if you want it split out.

PR #209 fixes a different defect in the same area (Textarea appending its defaults). No conflict. Textareas on this branch still render rows="5 4", so the new textarea test asserts a clean wrapper rather than an exact rows value.