You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
gate-40 flags the hidden file input gate-37 declared canonical — a finding that cannot be closed honestly #273
Two gates in the a11y band contradict each other on the same element
gate-37 (aria-hidden-focusable) was repaired in #251 on the finding that tabindex="-1" is the attribute that removes an element from the tab order, and that the canonical hidden-file-input pattern
<inputref="picker" type="file" :aria-hidden="true" tabindex="-1" @change="onPick"><NcButton@click="$refs.picker.click()">Upload a file</NcButton>
is correct — both of the gate's old remediations (drop aria-hidden, or drop tabindex="-1") regress accessibility.
gate-40 (form-label-association) flags that same element:
Measured 2026-08-08 at package sha cdfbd7a, as a control fixture in both a .vue app (opencatalogi) and a PHP-template app (nldesign) — gate-40's count rose by exactly 1 in each.
Why the finding cannot be closed honestly
aria-hidden="true" removes the element from the accessibility tree. An element that is not in the tree has no accessible name, so aria-label on it is inert — the only two ways to make gate-40 green are to add an attribute that does nothing, or to remove the aria-hidden that gate-37 requires. This is the same shape as gate-40's own NcCheckboxRadioSwitch finding (#251) and gate-43's spacer-<th> finding: a remediation that ships a change helping no one. check_table_headers.py already encodes the principle — it exempts a <th> with aria-hidden="true" because "a header with no accessible name names nothing".
Why I did not just fix it
Zero fleet occurrences today, so a fix has no measured benefit and some blinding risk:
nldesign's three gate-40 findings are the same shape but use style="display:none", not aria-hidden. Excluding display:none would be genuinely unsafe — a style can be toggled by JS, so the element may well be exposed at runtime.
Excluding literal, static aria-hidden="true" is defensible on the same grounds gate-43 already uses, but it widens the set of things gate-40 refuses to look at, and "tightening a gate creates its own false positives" cuts both ways.
Recording the measurement and the reasoning rather than guessing at the remedy.
Suggested fix, if taken
In scripts/lib/check_form_labels.py, exempt an input carrying literal or bound aria-hidden="true" — mirroring ARIA_HIDDEN_TRUE in check_table_headers.py — and ship it with the true positive it must not swallow: the same input withoutaria-hidden must still fail. Leave display:none alone.
Found while running the gates 34–44 acceptance test (#272).
Correction to the framing — an exemption keyed to aria-hidden would have matched NOTHING, and this is now fixed keyed to the spelling that actually occurs.
The measurement
Ran gate-40's helper full-tree across 21 fleet repos (the runner's own scope: src + templates + appinfo/templates, .vue/.php/.html). Of 471 findings:
attribute on the finding
count
aria-hidden="true"
0
style="display: none"
8
...of which type="file"
8
...of which anything else
0
So the suggested fix — "exempt an input carrying literal or bound aria-hidden="true", mirroring ARIA_HIDDEN_TRUE in check_table_headers.py" — would have been correct in principle and swallowed zero of the findings it was written for. This issue says nldesign's three "are the same shape but use style="display:none""; that is not a footnote, it is all eight fleet occurrences:
The principle is "not in the accessibility tree" — an element that is not in the tree has no accessible name, so aria-label on it is inert, and the only two remedies are a no-op or deleting the attribute that hides it. Both spellings are now recognised, but not on the same terms, because this issue is right that they are not equally static:
aria-hidden="true" (literal or :aria-hidden="true") — an authored declaration. Honoured on any input, on the same evidence gate-43 already uses for <th>.
display:none — a style, and "a style can be toggled by JS" stands. Honoured only for type="file", where hidden-and-clicked-programmatically is the canonical pattern (the visible, labelled trigger is a separate button) and a toggled-visible file input is not a real shape. All 8 occurrences are exactly this and 0 are anything else, so the restriction costs nothing today and is what stops the exemption becoming "style yourself invisible to silence the gate".
A dynamic:aria-hidden="someVar" is not accepted — its value is unknown at scan time.
Verification (NotInTheAccessibilityTree, 7 arms)
Each relaxation ships with the true positive it must not swallow:
arm
expectation
display:none file picker
0
:aria-hidden="true" file picker
0
same file input with the hiding attribute removed
1
display:none on type="text"
1 ← the abuse control
display:none on a <textarea>
1
:aria-hidden="isCollapsed" (dynamic)
1
aria-hidden="false"
1
All four FP arms fail on origin/main and pass with the fix; the TP arms pass in both directions.
Fleet effect
gate-40 549 → 462 across 21 repos (this change accounts for −8; the other −79 is #310). nldesign 3 → 0.
Activity
fix(a11y): name the three hidden file inputs (#255)
rubenvdlinde commentedon Aug 10, 2026
Correction to the framing — an exemption keyed to
aria-hiddenwould have matched NOTHING, and this is now fixed keyed to the spelling that actually occurs.The measurement
Ran gate-40's helper full-tree across 21 fleet repos (the runner's own scope:
src+templates+appinfo/templates,.vue/.php/.html). Of 471 findings:aria-hidden="true"style="display: none"type="file"So the suggested fix — "exempt an input carrying literal or bound
aria-hidden="true", mirroringARIA_HIDDEN_TRUEincheck_table_headers.py" — would have been correct in principle and swallowed zero of the findings it was written for. This issue says nldesign's three "are the same shape but usestyle="display:none""; that is not a footnote, it is all eight fleet occurrences:What was implemented
The principle is "not in the accessibility tree" — an element that is not in the tree has no accessible name, so
aria-labelon it is inert, and the only two remedies are a no-op or deleting the attribute that hides it. Both spellings are now recognised, but not on the same terms, because this issue is right that they are not equally static:aria-hidden="true"(literal or:aria-hidden="true") — an authored declaration. Honoured on any input, on the same evidence gate-43 already uses for<th>.display:none— a style, and "a style can be toggled by JS" stands. Honoured only fortype="file", where hidden-and-clicked-programmatically is the canonical pattern (the visible, labelled trigger is a separate button) and a toggled-visible file input is not a real shape. All 8 occurrences are exactly this and 0 are anything else, so the restriction costs nothing today and is what stops the exemption becoming "style yourself invisible to silence the gate".A dynamic
:aria-hidden="someVar"is not accepted — its value is unknown at scan time.Verification (
NotInTheAccessibilityTree, 7 arms)Each relaxation ships with the true positive it must not swallow:
display:nonefile picker:aria-hidden="true"file pickerdisplay:noneontype="text"display:noneon a<textarea>:aria-hidden="isCollapsed"(dynamic)aria-hidden="false"All four FP arms fail on
origin/mainand pass with the fix; the TP arms pass in both directions.Fleet effect
gate-40 549 → 462 across 21 repos (this change accounts for −8; the other −79 is #310). nldesign 3 → 0.
Closed by the same PR as #310.
Fixed in #329.
fix(gates): eight false-positive defects across gates 3/6/7/26/29/40/…