The Invisible Chips: an iOS 26 Bug Fixed by One Point of Padding
On the station list of Fuelvo’s iOS app, under the search field, there’s a row of chips: Open now and one per fuel type. It’s how you switch from diesel to petrol. On iOS 26, on most launches, that row wasn’t there.
It hadn’t been removed. The chips were laid out, they responded to taps, and the UI tests found them. They just weren’t drawn. I didn’t find it through a crash report or a failing test, but by looking at my own App Store screenshots: the row was blank in every screenshot of the list. Version 1.1.0 had passed App Review and was live with the bug.
The layout
The row is a horizontal ScrollView inside a header that the list pins under the navigation bar
with safeAreaInset:
List { /* stations */ }
.safeAreaInset(edge: .top, spacing: 0) {
VStack(spacing: 0) {
ScrollView(.horizontal, showsIndicators: false) {
HStack(spacing: FuelvoSpacing.sm) { /* Open now + one chip per fuel */ }
.padding(.horizontal, FuelvoSpacing.md)
}
}
.padding(.bottom, FuelvoSpacing.sm)
.background(.bar)
}
This is standard SwiftUI: a header that stays put while the list scrolls under it. It worked on iOS 18 and it works on iOS 27. It failed on iOS 26.5.
First fix: plausible, tested once, and wrong
iOS 26 introduced Liquid Glass and, with it, the scroll edge effect: content that scrolls under
a bar fades out. The chips’ row sat right under the bar, so the obvious suspect was that the system
was fading the whole row. iOS 26 has a modifier to opt out, scrollEdgeEffectHidden, and wrapping
the row in it made the chips show up on the next launch. The ticket (FUEA-123) was closed, with
a screenshot of before and after.
Later that day, preparing version 1.1.1, I asked a different question: on how many launches does it show up? I counted clean installs (delete the app, install it, launch it, check the row) on the iPhone 17 simulator with iOS 26.5:
| Launches where the chips were drawn | |
|---|---|
| Without the fix | 6 of 32 |
With scrollEdgeEffectHidden | 19 of 38 |
The fix had raised the rate from about one in five to about one in two. It hadn’t fixed anything: my first test had happened to land on a good launch. Another detail made the bug worse: a row that started blank stayed blank for the whole session.
Diagnosis by elimination
With a rate instead of a yes/no, each hypothesis could be tested properly: change one thing, then run enough clean installs to see whether the rate moved.
- The same row without a
ScrollView: drawn on 10 of 10 launches. So the scroll view was part of the problem, not the chips or the bar’s material. safeAreaBarinstead ofsafeAreaInset, hiding or hardening the edge effect on the list, dropping the.barbackground, and rebuilding the row: none of these moved the rate meaningfully.- Moving the row out of the
safeAreaInsetdrew it every time, but broke something else: after switching tabs, pulling the list down no longer revealed the search field. Not acceptable.
Looking at the scroll view’s frame showed what was going on. The scroll view’s frame didn’t start under the bar where it appeared to be. It started at the top of the screen, and the chips occupied the last 44 points of a frame that extended up under the navigation bar. A scroll view that touches the top safe area extends into it, and on iOS 26 what a scroll view holds in that region often isn’t drawn.
The fix
ScrollView(.horizontal, showsIndicators: false) {
HStack(spacing: FuelvoSpacing.sm) { /* … */ }
.padding(.horizontal, FuelvoSpacing.md)
}
// Not a margin: a scroll view that touches the top safe area stretches up under the
// navigation bar, and iOS 26 does not draw what a scroll view holds there. On most
// launches the chips were laid out, tappable and invisible (`FUEA-126`).
.padding(.top, 1)
One point of padding keeps the scroll view off that edge, so it no longer extends under the bar.
Result: 56 of 56 launches drawn, with the layout and every behaviour unchanged, the search
field included. The scrollEdgeEffectHidden wrapper was removed in the same commit, since it
wasn’t the cause.
The comment matters as much as the padding. Without it, a 1-point padding looks like a rounding error, and the next cleanup removes it.
The fix shipped in iOS 1.1.1, which Apple has approved. It’s been checked on the iOS 26.5 and 27.0 simulators. It hasn’t been checked on a physical iPhone or on iOS 18.x, and the commit says so.
What I took from it
- An intermittent bug needs a rate, not a screenshot. One good launch proved nothing about
FUEA-123. “Fixed” now means a number across many clean installs, both before and after the change. - UI tests check that elements exist, not that they’re visible. XCUITest saw the chips, could tap them, and the taps worked. Only a person looking at the screen, or a screenshot, could catch this.
- Turn the finding into a rule. The project’s guidance now says that a scroll view must not touch the top safe area under a bar, and that a fix for something that only happens on some launches is measured over many clean installs, never judged from one.
- Look at your store screenshots carefully. They’re the most-viewed images of your app, and in this case they were also the bug report.