Repository navigation
Close SLICE-014 and SLICE-015, and settle the two selector calls - #117
Merged
Merged
Conversation
Both slices are stamped Done, but not by ticking everything. Two items under SLICE-015 stay open and are named where they live rather than in the done list: press-fidelity renders at each scale factor, and the render half of the frame budget. One item is retired rather than skipped. The LCD-subpixel check belonged to the texture recipe that fed the mesh renderer, and the owner's gate kept the projective path, which rasterises no texture and samples nothing. There is no resampled glyph left to check. What did get measured is the arithmetic: a homography solve is 12.897us and a field sample 0.802us, so one deforming key costs 0.155 percent of a 120Hz frame. The mesh path it replaced measured 2999us per key per frame. Worth writing down because the gate that chose between them was decided on looks, and the cheap answer turns out to be about 230x cheaper as well. SLICE-014's own eyeball entry records that the owner found a breach of the guardrail that slice wrote for itself - never make a control's state less legible - and that nothing caught it because WCAG exempts disabled controls, so nothing had ever measured them. The two selector calls are settled with measurements rather than taste. CTDB sharing stays a window: the Settings row is 455 wide, its control gets 296, and three keys plus that row label need about 540. Queue action becomes a bank, because the earlier reasoning was wrong on a fact - it is a toolbar row, not a table row, and its two options are short. The layout test now sweeps every page carrying a bank rather than just Rip. A page left unvisited is a page left unchecked, and the clipping it looks for is invisible at page level.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Slices closed honestly, not by ticking
Both are stamped Done. Two SLICE-015 items stay open and named where they live rather than in the done list: press-fidelity renders at each scale factor, and the render half of the frame budget.
One item is retired rather than skipped - the LCD-subpixel check belonged to the texture recipe feeding the mesh renderer, and the gate kept the projective path, which rasterises no texture and samples nothing. There is no resampled glyph left to check.
What did get measured: a homography solve is 12.897 us, a field sample 0.802 us, so one deforming key costs 0.155% of a 120 Hz frame. The mesh path it replaced measured 2999 us/key/frame - 36% of the same budget. The gate chose the cheap path on looks; it turns out to be ~230x cheaper too.
SLICE-014's eyeball entry records that the owner found a breach of the guardrail that slice wrote for itself (never make a control's state less legible), and that nothing caught it because WCAG exempts disabled controls, so nothing had ever measured them.
The two selector calls, settled on measurements
CTDB sharing stays a window. The Settings row is 455 DIP, its control gets 296, and three keys (~352) plus the row label (~190) need ~540. It would fit with the label stacked above, but that is the only row on the page that would break the label-left rhythm. Shortening to Ask/Always/Never would work - not taken, because it is a consent control,
settings.md:34quotes the strings, and the long labels are the ones that still say what they mean when read without the row label. If you want it, that is three strings, one manual line and an accessible name - five minutes, reversible.Queue action becomes a bank. My earlier reasoning was wrong on a fact: it is not in a table row, it is a
LastChildFill="False"toolbar where a wider control just pushes the next along. Options areVerifyandConvert- the two-short-label rocker case.Test
The layout sweep now visits every page carrying a bank, not just Rip. A page left unvisited is a page left unchecked, and the clipping it looks for is invisible at page level. 194 green.