The 4x12 rectangular stretch from v1.2.3 read as "weird". Requirement
is square cells + comfortable side padding. Given the 53-week hard
constraint, those two requirements together pick the cell size:
53 * (size + gap) + leftPad + rightPad = 340
Candidates audited:
6 x 6, gap 1 → 371 wide, overflows the frame
5 x 5, gap 1 → 318 wide, only 11 px of total side padding ("very close")
5 x 5, gap 0 → 265 wide, cells touch (need a stroke to fake a gap)
4 x 4, gap 1 → 265 wide, 30 + 45 px gutters, real 1 px gaps ✓
3 x 3, gap 2 → 265 wide, cells become pinhead-sized
4 x 4 with a 1 px gap is the largest square that keeps breathing room
on both sides without any rendering trickery. Grid is 35 px tall;
there's leftover vertical space on the card but short beats "stretched
horizontal bands" visually. Bump topPad from 62 → 70 to offset the
grid slightly down from the title and reduce that apparent emptiness.
Legend swatches also revert to the same cellSize so the Less/More
bar matches the grid visually again.
Cells were 4 × 4 with 1 px gap: the grid used 35 of 140 available
vertical pixels (25 %) and ~80 px of dead space below. Widening the
cells isn't an option — 53 weeks already consume every horizontal
pixel past the weekday-label gutter. So stretch vertically instead:
cellW = 4 (unchanged) cellH = 12 cellGap = 1
Grid footprint is now 265 × 91 px inside a 340 × 200 frame, with a 32
px gap below the grid for the legend. Each weekday reads as a distinct
horizontal band instead of a cramped postage-stamp row.
Legend keeps 8 × 8 SQUARE swatches so the "Less ▢▢▢▢▢ More" row is
still recognisable as an intensity legend rather than a stretched echo
of the data cells. Label baseline offset switches from `cellSize - 1`
to `cellH - 3` so "Mon"/"Wed"/"Fri" sit visually centred in the taller
rows.
Two follow-ups:
1. "Top 7 languages" counts rows the reader sees, not "7 named + Other".
Revert collapseOther to the original semantic: when the input has more
than n entries, keep in[:n-1] named and fold the tail into a single
"Other" row — 7 rows total with topN=7. Input with ≤7 real languages
still passes through unchanged. Test and adversarial-profile updated
to pin the cap (6 named + Other, Shell/Kotlin/Java collapse).
2. Contributions heatmap grid was touching the card right edge
(leftPad 22 + 53*6 = 340 exactly). Shrink cells from 5×5 to 4×4 and
bump leftPad to 30, giving 45 px of right gutter. Grid no longer
reads as bleeding into the frame border; weekday label column also
gets a few more pixels of breathing room on the left.
With topN=7 the previous collapseOther kept only the first 6 entries
and added "Other" as the 7th row. A user expecting to see 7 actual
languages in the legend saw six named languages plus "Other" — the
exact complaint just raised about the profile repo's donut.
Flip the semantic: the "top N" slots are reserved for real languages,
and "Other" is an extra row when (and only when) there's a non-zero
tail past the Nth entry. Topologically that means up to 8 legend
rows — still fits the card frame (row 8 text baseline at y=195, card
height 200).
- TestDonutTopSevenPlusOther pins the new contract with a 9-language
input.
- adversarialProfile in TestCardsFitFrame bumped to 9 languages so
the stress test exercises the 8-row legend geometry.
- design-guidelines: the donut row re-reads "Up to 7 named languages,
plus an 'Other' row when the tail is non-zero (8 rows max)".
project-roadmap.md went from a 146-line phase-by-phase history to a
48-line focused view: what's planned, what's out of scope. Completed
work is already in git log + GitHub Releases — the doc re-telling it
was the thing most likely to rot and least likely to be read.
project-overview-pdr.md: "Open questions" section dropped its stale
bullet list and now just points at project-roadmap.md (single source of
truth for planned work).
code-standards.md: drop the ".claude/ directory" commit rule — that's
a per-user workflow detail, not a project-level standard. Docs are for
users of the CLI/Action and coworkers of this repo, nothing else.
The repo is already listed on the Marketplace as ghstats-cards, so the
multi-paragraph "open the release page, tick the checkbox, re-publish"
instruction was stale. Replace with a one-line note pointing at the
existing listing and stating that new releases inherit visibility
automatically.
The README previously only LINKED to the demo gallery. Pull the dracula
theme inline so a reader can scan every card type + the LY/AT pairs
without leaving the page, and use the demo gallery for the other 64
themes. SVGs resolve against demo/dracula/ which the demo workflow keeps
fresh on every push to main.
Several recent code changes hadn't propagated to the docs:
- design-guidelines
* Card frame title row: document the 11–15 px auto-shrink (not a flat
15 px anymore).
* Donut Top-N: already 7 (updated earlier).
* Bar-chart section renamed to cover weekday + by-year too; document
the peak-vs-dim highlight convention and the niceTicks yMax ≥ max
invariant.
* Add Heatmap / Stat-column (streak) / List (top-starred) card
sections — they were missing entirely.
* Rewrite "Text overflow" from "we don't truncate" to the current
truth (truncate helper, formatTick abbreviations).
* Replace dangling `truncateName` reference with `truncate`.
- code-standards
* SVG output standards: call out truncate, formatTick abbreviation,
header auto-fit so the card-review gate reflects what the renderers
actually do.
- codebase-summary
* Layout tree: svg.go / axis.go comments list the helpers they now
contain; productive.go notes the weekday histogram.
* demo/ tree shows the index vs per-theme split.
* Data-flow diagram includes Weekday in the productive pass.
* Test coverage row lists TestCardsFitFrame / TestFitTitleFontSize /
TestNiceTicksCoversMax — the new invariant guards.
- system-architecture
* Shared primitives list adds renderWeekday and renderHeatmap; donut
blurb updated to "top 7".
* New "Chart-geometry invariants" block documents niceTicks ceiling,
formatTick abbreviation, header auto-fit.
- project-roadmap
* Phase 7.5 bullet updated to describe the index + per-theme demo
split (was single-README TOC).
Legend at x=20-155 fits 7 rows (y=55 to y=175); donut at cx=250, cy=110
is unaffected. Update the design-guidelines table to match. Stress test
still passes — the legend column never leaves its left-side gutter.
Loading demo/README.md was unusable — 65 themes × 15 SVGs = 975 images
in one page, and reloading made the tab lag for 10+ seconds. Restructure:
- demo/README.md shrinks to an index: one bullet per theme linking to
that theme's page. Zero images.
- demo/<theme>/README.md (new) embeds the 15 SVGs for that theme only,
grouped so the last-year / all-time variants sit side-by-side in HTML
tables. Readers compare LY vs AT without scrolling back and forth,
and fetch ~15 images instead of 975.
Per-theme page layout:
## At-a-glance profile · stats · streak (each full-width)
## Repos top-starred · repos-per-language
## Contributions heatmap · by-year, then
monthly shape: LY | AT (paired)
## When you commit hour-of-day: LY | AT
day-of-week: LY | AT
## What you commit language share: LY | AT
pair() and single() bash helpers skip missing SVGs so a future
partial-render doesn't emit dangling <img> tags.
The title fitter was correct but the loop-down-from-maxFont form hid
its intent. Switch to the direct expression:
ideal = floor(budget / (chars * 0.6))
clamp to [11, 15]
Same answer for every title; easier to verify at a glance. Hoist the
four constants (leftInset=20, rightSafety=4, minFont=11, maxFont=15,
charRatio=0.6) to package scope so the new unit test can reference
them without re-declaring.
Utilization audit for realistic dracula titles (width=340, budget=316):
Stats (5) → 15 px (14 %)
Top Starred Repos (17) → 15 px (48 %)
Most Commit Language (all time) (31) → 15 px (91 %)
Commits by Hour (last year, UTC+7.00) (37) → 14 px (98 %)
Commits by Weekday (last year, UTC+7.00) (40) → 13 px (99 %)
Commits by Weekday (last year, UTC+12.75) (41) → 12 px (93 %)
200-char pathological → 11 px (floor)
TestFitTitleFontSize pins this table so a future charRatio tweak can't
silently regress any real title. TestCardsFitFrame (the end-to-end
check) plus this unit test now cover both the geometry and the picked-
font-size paths.
When niceTicks picks a step that doesn't divide the data max, the last
returned tick was the highest step multiple ≤ max. Callers used it as
yMax, so any data point > lastTick rendered a bar > chartH that poked
above the chart top into the title area.
Concrete case from the dracula demo: max=625 with step=100 → ticks
[0, 100, 200, 300, 400, 500, 600], yMax=600, bar height for 625 =
110*(625/600) = 114.58 — 4.58 px past the chart top and right against
the card title.
Fix in niceTicks itself: round the top tick UP to the next step multiple
(`last = ceil(max/step) * step`), so 625 yields [..., 600, 700] and the
same 625 bar lands at 110*(625/700) ≈ 98.2 px, with a clean 12 px gap
to the title.
This is the stable answer to title-vs-bar collision: regardless of
which weekday (or year, or month, or hour) holds the peak, the chart
headroom is built into the axis instead of leaned on per-card. The
title auto-shrink from the previous fix still applies — that's for
literal text width, an orthogonal problem.
Add TestNiceTicksCoversMax covering the cases (625, 99, 101, 7, 49,
999, 1001) that would have silently regressed before.
productive-weekday's "Commits by Weekday (last year, UTC+7.00)" is 40
chars — at font-size 15 that renders ≈360 px wide, past the 340 px
frame. Before the previous stress-test run missed it because the
fixture left UTCOffsetLabel blank, so the title was the short
"Commits by Weekday (last year)" form that happens to fit.
Fix at the render layer instead of rewriting every long title:
header() now picks the largest font between 11 and 15 px at which the
title still fits in 316 px (card width − 20 left inset − 4 right
safety). 40-char titles land at 13 px; ≤ 35-char titles still render at
15 px as before. Readability floor is 11 px — far from the point at
which text becomes illegible, but a hard floor nevertheless.
Stress-test fixture now populates UTCOffsetLabel with "UTC+12.75"
(half-hour / quarter-hour zones make the widest title) so this class
of bug surfaces automatically from here on.
The stress test caught real overflows from profiles the author doesn't
have:
- Y-axis tick labels were raw integers. A user with 10,000+ yearly
contributions or 1000+ monthly commits would render "10000" / "1000000"
at text-anchor="end" against a ~28 px gutter — the digits spilled
leftward past x=0. formatTick now abbreviates: 999→"999", 1500→"1.5k",
12345→"12k", 1234567→"1.2M". No label exceeds 4 chars, all fit the
gutter for every card that uses niceTicks.
- Profile details dumped Company / Location / Website / title verbatim,
which works for the author (VNG, Ho Chi Minh, miti99.com) but not for
40-char strings. Each row truncates at 40 runes; the title truncates at
34. Uses a rune-aware truncate() helper hoisted out of top-starred-repos
into svg.go so every list-style card can share it.
- Streak date range collapses to a single-year form ("Jan 2 — Dec 31")
when start.Year() == end.Year() and to "YYYY — YYYY" across years. The
previous "Jan 2 — Dec 31, 2025" format at 10 px × 21 chars pushed past
the ~113 px column width.
The TestCardsFitFrame stress test was reading text-anchor and font-size
with a non-greedy regex that missed attributes whose position varied. It
now parses the opening <text> tag as a block and extracts each attribute
with its own regex, so text-anchor="end" / "middle" elements are no
longer false negatives. The check also estimates rendered width
(0.6 × font-size × len) and asserts the implied left/right edges stay in
the frame — catching exactly the class of bug the axis-tick case
represents.
Three cards overflowed the 340×200 frame for realistic profile data:
- contributions-heatmap: the classic case — 53 weeks at 9px cellSize + 2px
gap pushed the grid out to x≈611. Shrink to cellSize=5, cellGap=1 so
leftPad(22) + 53*6 = 340 (exact fit). Drop month labels within 20 px of
the right edge so "Dec"/"Apr" can't stick past the frame.
- streak: the third column rendered "N / M" at font-size 28, centered at
x=282. For 4+ digit totals (e.g. 584 / 3031) the text extended to x≈347.
Refactor to show the active-days integer by itself in the big slot and
push "of N total (P%)" into the small detail line that the other two
columns already use.
- top-starred-repos: the per-row star icon sat at x=306 while the right-
anchored number ended at x=334, so 5+ digit star counts collided with
the icon. Drop the icon (card title already says "Top Starred Repos"),
emit the count as "N ★", right-anchor at x=334 with a 6 px safety gap.
Add a new TestCardsFitFrame stress test that renders every card against an
adversarial profile (10-digit counts, 40-char names, 20 active years,
53-week span) and asserts every positional attribute stays inside the
frame. This is the automated half of the new "fit-the-frame invariant"
added to docs/design-guidelines.md + a pre-release review checklist in
docs/code-standards.md.
Bug reports will still surface text-overflow cases that the coordinate
check can't see (a text-anchor="middle" element has a single x attribute
but renders outward), so the docs also spell out the human-review step:
render dracula against tiny/typical/adversarial fixtures before release.
The hardcoded cards[] array in demo.yml still listed the original 9,
so the S-tier additions (heatmap, streak, by-year, weekday pair,
top-starred) were rendered into demo/<theme>/ but never referenced in
demo/README.md. Expand the array to the full 15 in README card-table
order.
The dracula sample under output/dracula/ was a reference render for the
README. Now that demo/ auto-generates every card × every theme on each
push, the sample is redundant — and keeping it trackers means every
behavioral tweak shows up as a diff in committed SVGs.
- Delete output/dracula/*.svg.
- .gitignore: collapse `output/*` + `!output/dracula/` to a plain `output/`.
- demo.yml: drop `output/**` from paths-ignore (no longer part of any push).
- README / docs: point readers at demo/ instead of output/dracula/.
Five new cards, all derived from data FetchProductive / FetchProfile already
pull, so zero additional API calls:
- contributions-heatmap: 7×53 calendar grid with a 5-bucket intensity ramp
mixed from each theme's Background→Accent so palettes with no dedicated
heat ramp still render sensibly.
- streak: current streak, longest streak with date ranges, active/total days.
- contributions-by-year: one bar per active year, peak year highlighted.
- productive-weekday + -all-time: 7-bar day-of-week mirror of the hour-of-day
cards; FetchProductive now also fills Weekday / WeekdayAllTime histograms
during the same commit-history pass.
- top-starred-repos: top 5 owned non-fork repos by stargazer count; threads
Stars through RepoInfo.
Card count: 9 → 14. Registered in allCards grouped by recency (last-year
block, then all-time block). Render test extended to cover all new files
and realistic daily-series inputs.
- README points to demo/ as an auto-generated theme preview.
- Bump "60+ / 61 themes" references to the actual 65.
- Record Phase 7.5 in roadmap and list demo.yml + demo/ in layout.
Runs on push to main (skipping its own demo/output/markdown commits),
builds the ghstats binary, renders every theme for the repo owner,
and writes a demo/README.md gallery back to main.
Adds an update-major-tag job that runs after test/docker/binaries all
pass. Extracts the major component from the pushed tag (v1.2.3 → v1)
and force-moves that tag to the current commit, then pushes.
Consumers using `uses: tiennm99/ghstats@v1` now pick up each new
minor/patch release automatically without a workflow edit on their
end. The v1 tag only moves if every artifact job succeeded, so it
never points at a commit whose binaries or Docker image failed to
publish.
This reverts commit 399a3dc. Repo stays as tiennm99/ghstats; the
Marketplace display name ("ghstats-cards" in action.yml) is the only
place the new name remains, since that field requires uniqueness on
the Marketplace.
plans/reports/* were added alongside the rename in the same commit
and are preserved by not deleting them in this revert.
Matches the Marketplace name; repo is being renamed in lockstep.
- go.mod module path: github.com/tiennm99/ghstats →
github.com/tiennm99/ghstats-cards
- Import paths across every .go file updated.
- README badges, install snippets, and the 'go install' line point
to the new URL/path.
- docs/deployment-guide.md workflow template, Docker image path, and
release edit URL updated.
Breaking for consumers pinned to the old URL; they need to swap
tiennm99/ghstats → tiennm99/ghstats-cards in workflows and switch
Docker pulls to ghcr.io/tiennm99/ghstats-cards. GitHub's HTTP
redirect covers git clones but GHCR does NOT redirect — users must
update image URIs manually.
- Marketplace / Release / License badges at the top.
- Explicit Marketplace + source links under the intro so visitors
arriving from the Marketplace page can jump to the repo and vice
versa.
Also set repo topics via `gh repo edit`: ghstats, ghstats-cards,
github-action, github-marketplace, profile-readme, profile-stats,
stats-cards, svg-cards, go, github-stats.
"ghstats" is already taken by an existing action/user/org on the
GitHub Marketplace, blocking publication. Rename the action's Marketplace
display name to ghstats-cards (still more descriptive of what it
produces). The repo URL, Docker image tag, and import path stay
tiennm99/ghstats — only the Marketplace listing label changes.
- Fetcher signatures across codebase-summary and system-architecture
now show the ctx-first arguments and document the rate-limit retry
loop in Client.query.
- Attribution pseudo-code hoists the per-repo total out of the commit
loop to match the current implementation (I6).
- Failure-modes table enumerates primary rate-limit retry, per-year
nil-user warn, and -timeout / Ctrl-C cancellation.
- design-guidelines notes the single-slice donut special case.
- deployment-guide's release section documents the new test gate and
the SHA-pinned Docker/GHA actions; troubleshooting adds the
rate-limit-reset-too-long error. Rate-limit section describes the
sleep-and-retry policy and -timeout flag.
- project-roadmap records Phase 6 (code-review remediation) as done,
renumbers later planned phases, links the new review report.
- I4 — TestRenderAll now seeds Name/Company with XML-significant chars
that actually hit the render pipeline via cardTitle(). Previous test
checked Bio which is no longer rendered, so the assertion was vacuous.
- New TestDonutSingleSlice guards against the I1 empty-arc regression:
asserts the single-slice path emits <circle> primitives and not the
degenerate A-command path.
- New TestDonutEmpty covers the zero-stats fallback.
- New TestUTCOffsetLabel pins the UTC±N.NN format across UTC,
Asia/Saigon, Asia/Kolkata (half-hour), Asia/Kathmandu (quarter-hour).
Catches %+.2f regressions.
- N2 — Client.query, FetchProfile, FetchContributionsAllTime, and
FetchProductive now take context.Context. main.go builds a root
context with a -timeout flag (default 30m) and cancels on
SIGINT/SIGTERM so Ctrl-C aborts in-flight pagination cleanly.
- N6 — on HTTP 429 or 403 with X-RateLimit-Remaining: 0, the client
now parses Retry-After / X-RateLimit-Reset, sleeps (capped at 5
minutes), and retries once. Longer reset windows surface as errors
so the caller can reschedule the Action rather than burn 55 min of
runner time idle.
N8 — the field is set from contributionsCollection.contributionCalendar,
a rolling last-year window, not a lifetime total. Rename + updated
comment makes the semantics explicit so future readers don't confuse
it with TotalCommitsAllTime.
- I1 — donut chart with a single slice (100%) now renders via two
concentric <circle> elements instead of a degenerate SVG arc that
drew nothing. Reproduced with a standalone probe; regression test
added separately.
- I2 — FetchContributionsAllTime logs a warn to stderr when a year's
query returns nil user data so callers notice partial results
instead of rendering an empty all-time card silently.
- I6 — attributeCommit() receives the repo's byte total precomputed
once per repo rather than re-summing language edges for every
commit in the inner loop.
tiennm99andGitHub