ADFA-5527: Follow the system font scale in the metrics charts, up to a ceiling

[davidschachterADFA]

What

The metrics charts now follow the system font scale, up to a ceiling of 1.5.

MPAndroidChart sizes its text in dp, not sp, so nothing the charts drew responded to the font scale at all — a user who asked for larger text got it everywhere in the IDE except inside these plots, where the text is already the smallest on the screen. Measured before this change, on the same text in two screenshots normalised to one scale:

Text	1.0	2.0
x-axis label (-54s), drawn by the chart	19 px	20 px
legend text, drawn by the chart	20 px	21 px
carousel title, an ordinary TextView beside them	36 px	54 px

Why a ceiling, and not the full scale

This is option 1 of the four written up on ADFA-5527. A plot is dense by nature and the strip is a fixed editor_mem_usage_view_height of 248dp. At the platform's full 2.0 the axis labels collide with each other and the eight staggered annotation rows overlap, so honouring the scale literally would make the chart less readable rather than more. 1.5 gives most of the benefit and keeps the plot legible at the extreme.

Growing the text was not enough on its own

The first version of this change scaled the text and left the label count alone. The result was worse, not better: the memory page's nine value labels went from 29 px apart to 6 px — bigger text, crowded axis. The count now falls as the text grows, from six at scale 1 to four at the ceiling, and is applied as a hint rather than a command so that granularity still has the last word. That matters on the temperature axis, which ADFA-5499 pinned to whole degrees.

The annotation rows scale too. Eight rows sized for scale-1 text would have overlapped exactly when the labels grew, which is what the staggering in ADFA-5509 exists to prevent.

Verified on device, not by eye

Pixel 6 Pro, measured from screenshots rather than judged:

	font scale 1.0	font scale 2.0
Memory page value labels	9 labels, min gap 31 px	5 labels, min gap 28 px
Temperature axis	—	5 labels, whole degrees, no repeats, min gap 97 px
Power axis	—	5 labels, whole watts, no repeats, min gap 28 px

Same readability, visibly larger text. The interaction I was most worried about — the label-count hint fighting the whole-degree granularity on the temperature axis — did not materialise: granularity wins, as the code comment claims.

Six unit tests pin the policy. Three of them fail against the unfixed code; the other three are guards (no change at scale 1, no shrinking below scale 1, eight rows still fit at the ceiling). The annotation-row test asserts the yOffset of a drawn marker rather than calling the helper, because an earlier version called the helper directly and passed even with the renderer still using the unscaled constant.

Scope

Not part of the metrics carousel stack, deliberately. It touches the shared renderer and the annotation geometry, and five reviewed PRs (#1784, #1785, #1790, #1791, #1792) are waiting on approval — adding it there would invalidate all of them to fix something that predates them. The charts have behaved this way since the memory chart was first added; the carousel work made it reach three pages instead of one.

Branched off feature/ADFA-5509-build-annotations rather than stage, because MetricsChartRenderer does not exist on stage yet. This PR is blocked by that whole stack landing.

Review fix: the test that justifies the ceiling now measures

eight annotation rows still fit the plot at the ceiling compared eight rows against a hand-picked PLOT_HEIGHT_DP = 150f, with a comment admitting the number was a guess. A guess pins nothing — no layout change could move it, so the test could not tell a shorter strip or a taller title row from a safe one, which is the whole risk a font-scale ticket carries.

It measures the whole way down now, with nothing allowed for by hand:

• the strip's height is editor_mem_usage_view_height, read from resources;
• the pager's share of it comes from a real measure and layout of the real strip at 360dp wide, so the title row growing at 2.0 counts against it;
• the plot's share of that is the content rect a real item_metrics_chart page reports after a real renderer has put its legend and axis on it.

Raising MAX_TEXT_SCALE to 2.0 now fails it — 192dp of rows against a 169dp plot — which is the claim the test makes about why the ceiling is 1.5. It passed at 2.0 before.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QeW3M24yD6HNhEnbuNW7Hz

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}