Repository navigation
feat(benchmarks): group runs by workflow and hardware, one comparison per group - #2005
Closed
mariana953 wants to merge 20 commits into
Closed
mariana953 wants to merge 20 commits into
mariana953 wants to merge 20 commits into
Conversation
The Comparison title moves inside its card, styled like the other section labels, and the two Properties pickers become icon buttons with the label kept for screen readers, so the comparison toolbar fits one row. Matches the New UI proposal's Benchmarks page. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
|
PabloWiedemann
force-pushed
the
PabloWiedemann/sidebar-tabs-assets
branch
from
November 10, 2026 21:01
a3c501f to
569fb3a
Compare
Drop the card around the runs library and the Comparison section: filters, refresh, search and columns sit in one toolbar row, the table is a plain hairline list without the forced horizontal scroll, and Select folder moves to the page header. Empty states get a quiet dashed placeholder and row delete buttons stay muted until hovered. BrandedPageHeader gains an optional actions slot. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Series colours follow the platform proto palette (no yellow), the baseline marker and drop target use the neutral foreground, checked rows use a white check and a soft fill instead of a coloured stripe, and tables, the chart and empty states use fills instead of strokes. Tests updated for the new hex values. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… panel Use the display title with no subtitle and the same page column as Builds and Apps. The comparison title and its controls share one row in a single panel, the chart comes before the metric table, and the best cell is a neutral fill instead of green. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Use the design system card surface (bg-card) so Benchmarks matches the other pages. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One page, one card per workflow and hardware. Each card names the fastest build, shows a bar per session for the selected metric with its range, and a Vs column against the fastest run or a chosen baseline. Row menu sets the baseline, renames or deletes; selected rows export through a floating bar. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
Workflow name as a sentence-case title, a smaller summary with the lead as a percentage, row controls in the card gutter so rows align with the title, no axis row, softer bars. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
Filters on the left; refresh, Compare by, columns and search on the right, search last at 220px. Title without a subtitle, as on the other pages. The per-group summary sentence is gone (the Fastest pill and the Vs column carry it). Wider gutter for the row controls. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
…er app tables Filters and Compare by use BaseSelect with a new optional label prefix, so every dropdown is one component and options never wrap. Table head, row hairlines and hover follow the shared table criteria. Select folder is md. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The five toolbar dropdowns use the design system Menu with radio items, the same one Apps uses for Filter and View, so size, spacing and margins match. BaseSelect goes back to its original shape. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
Collapsed groups show as compact rows in a card with tight spacing, and every group header shares the same inset so titles line up. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
Same centered pattern as the Apps empty state: heading, one muted line and at most one white action, with no card around it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mariana953
pushed a commit
to mariana953/Comfy-Desktop
that referenced
this pull request
Nov 11, 2026
…windows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drop the custom radius, padding and item overrides so every menu on the page renders with the design system menu, like the filter dropdowns. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
5 of 8 tasks
Contributor
Author
Chroma01
pushed a commit
to Chroma01/Comfy-Desktop
that referenced
this pull request
Nov 13, 2026
… per group (Comfy-Org#2015) * feat(benchmarks): group runs by workflow and hardware, one comparison per group Same change as Comfy-Org#2005, now based on main instead of the closed Comfy-Org#1981. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * style: page title matches the Builds title on main Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): compare against the whole group, split machines, fix reordering and exports Keeps Mar's grouped layout and rebuilds the logic behind it. - Runs group by workflow and machine profile (GPU, CPU, memory, OS), read from each run's results.json in main. Two PCs with the same GPU no longer share a table; the label adds the CPU (then memory, then OS) only when needed to tell machines apart. - Fastest, baseline, bar scale and best-value highlights come from every run in the group, not just the rows a filter leaves visible. When the reference row is hidden, the card says which session it is compared with. A chosen baseline survives filters. - "Only one build ran this workflow here" now counts builds in the whole group, so it no longer appears on every card when a Build filter or search is active. - Manual order is stored per group over all of its runs, so rows hidden by a filter, runs added after a Performance Test and runs from a newly selected folder can all be moved. Dropping onto a lower row places the row after it, so the bottom slot is reachable. Alt+Up/Down join Alt+Left/Right. Compare by no longer has a Manual option; a header click goes back to sorted order. - Selection count and export only include runs that pass the filters. Relative speed is exported only when the export is one group with a chosen baseline. Series colours follow export order. Reloading clears a stale export error. - Switching folders resets baselines, manual order and collapsed cards. - Session names rename in place with the shared inline-edit composable; the row menu's Rename focuses the name after the menu closes. - The columns picker is the design-system Menu with checkbox items, replacing the details element and its document-level listener. - Bars are focusable with an aria-label listing all four durations, so the tooltip works from the keyboard. - Grouping and comparison live in lib/benchmarkGroups.ts; each card is views/benchmarks/BenchmarkGroupCard.vue. * feat(benchmarks): group by choice, responsive table, column reorder and baseline flag Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): keep the range header on one line in narrow windows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): keep the toolbar on one row and fold search into an icon when narrow Filters sit on the left and wrap as a group; refresh, search and columns stay on the right. Below 720px of toolbar width the search becomes an icon button that opens the field in place, focused, and folds back on Escape or blur when empty. Select folder stays beside the page title. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): list every table column in the columns picker, uppercase its title The picker now shows the columns the table always has. Session, the Compare by value and Vs baseline are checked and locked; Workflow and the range bar can be hidden like any other column, and the Compare by metric no longer shows twice. The menu title uses the uppercase section label style. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): even toolbar rows at every width One row when everything fits. Below that, the data filters (workspace, build, hardware, workflow) take the first row and the view options (group by, compare by) share the second with refresh, search and columns. In narrow windows the search row goes on top and the six menus form an even two-column grid, so no menu is left alone on a row. Search is always visible, so the fold-to-icon toggle and its string are removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(benchmarks): announce Alt+Left/Right on column headers and keep their focus Column headers reused the row hint, which tells you to press Alt+Up or Alt+Down, in their tooltip and screen reader description. Columns move with Alt+Left/Right, so they now get their own hint. Moving a column to the left with the keyboard moved the focused header in the DOM, and Chromium dropped focus to the page, so the next Alt+arrow did nothing. The header button gets focus back after the move, as the row grips already do. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Benjamin Lu <benjaminlu1107@gmail.com>
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
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.
ELI5
Benchmarks used to be a big spreadsheet of runs plus a second table you had to fill by ticking rows. Now each workflow on each machine gets its own card with one table and one bar per session, and the fastest build is marked right in the row.
Summary
Menuwith radio items, the same one Apps uses for Filter and View, so size, spacing and margins match. The title has no subtitle, like the other pages.No functionality removed
Judgement calls
Before / after
Before is the base branch (#1981). After is this PR. Seeded data, 1320 px window. Screenshots use main's gray palette (#1996) applied locally: this branch is stacked on #1981, which predates it. The branch does not change the palette, so on main it renders gray.
Test plan
🤖 Generated with Claude Code