Analysis - FEATURE - Redraw every Analysis plot with Makie, loaded as a package extension

.github/workflows/julia-test-version.yaml

@@ -213,7 +213,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        group: [core, api, errorfields]
+        group: [core, api, errorfields, plots]
     env:
       GPEC_LOCAL_CACHE_DIR: ${{ vars.GPEC_LOCAL_CACHE_DIR }}
       JULIA_CPU_TARGET: 'generic'
@@ -293,11 +293,30 @@ jobs:
         env:
           CACHE_KEY: ${{ needs.precompile.outputs.gpec-key }}
 
+      - name: Update the package registry
+        if: matrix.group != 'plots'
+        run: julia -e 'using Pkg; Pkg.Registry.update()'
+
       # Level 1 can reuse the higher-optimization production package images.
       - name: Run tests
+        if: matrix.group != 'plots'
         uses: julia-actions/julia-runtest@v1
         with:
           coverage: false
           check_bounds: 'yes'
           optimize: '1'
           allow_reresolve: false
+
+      # The plots draw with Makie through GPEC's package extension, so Makie stays out of the package's own
+      # dependencies; this group loads CairoMakie from a separate environment stacked onto LOAD_PATH.
+      - name: Add the plotting backend
+        if: matrix.group == 'plots'
+        run: |
+          set -euo pipefail
+          backend="$RUNNER_TEMP/plots-backend"
+          julia --project="$backend" -e 'using Pkg; Pkg.add("CairoMakie")'
+          echo "JULIA_LOAD_PATH=@:$backend:@stdlib" >> "$GITHUB_ENV"
+
+      - name: Run plot tests
+        if: matrix.group == 'plots'
+        run: julia --project=. --check-bounds=yes -O1 test/runtests.jl

.github/workflows/test.yaml

@@ -61,7 +61,7 @@ jobs:
           # trigger a run, so new untested Julia (benchmarks, docs scripts, a
           # future tools/ directory) is excluded by default rather than by
           # remembering to exclude it.
-          tested='(^src/|^test/|^examples/|^ci/manifests/|^ci/local-julia-cache\.sh$|^Project\.toml$|^\.github/workflows/(test|julia-test-version)\.yaml$)'
+          tested='(^src/|^ext/|^test/|^examples/|^ci/manifests/|^ci/local-julia-cache\.sh$|^Project\.toml$|^\.github/workflows/(test|julia-test-version)\.yaml$)'
 
           relevant=$(printf '%s\n' "$files" | grep -E "$tested" || true)
 

CLAUDE.md

@@ -121,13 +121,7 @@ Roster, the recommended review pipeline (physics fidelity → readability → pe
 - **Default output**: `gpec.h5` (previously `euler.h5` in older versions)
 
 ### Plotting
-Spectrum-plot (`seriestype=:steppre` / `step_series`) and figure-saving conventions are in **[`docs/development/plotting.md`](docs/development/plotting.md)** — read before writing any plotting code.
-
-**Adding a plot.** Every plot in `src/Analysis/` takes a result in memory or a `gpec.h5` path and is registered in `src/Analysis/Registry.jl`, which `available_plots` and the REPL summaries read:
-- Read the data through one `_x_data(src)` method per source (the result's fields, and the same datasets from the file), and draw from the returned NamedTuple only, so both sources go through one plotting body.
-- Start the public `plot_x(src; ...)` with `_require(plot_x, src)`, which throws the registry's reason when the source lacks the data; never draw a "No data" placeholder or return `nothing`. Composite summaries skip the panels they cannot draw (`_can`) and throw only when none can.
-- Call `register!(plot_x, sources, needs, description)` after the definition. `needs` names capabilities; a new one gets a `capabilities` predicate for each result type that can hold it, a dataset in `_H5_CAPABILITIES`, and a `REASONS` sentence saying what to change in the run.
-- `test/runtests_analysis_plots.jl` checks every registered plot from both sources automatically; add an extractor to its list when the plot reads new data.
+Figures are drawn with Makie through the package extension `ext/GPECMakieExt/`. Read **[`docs/development/plotting.md`](docs/development/plotting.md)** before writing any plotting code: it has the figure conventions and the guide to adding an Analysis plot. In short, Analysis plots are added sparingly (frequent everyday-workflow figures only, approved by a human), and each is registered, follows the shared theme, carries little text, and throws an `ArgumentError` when its data are absent. Never offer to add an Analysis plot unprompted.
 
 **REPL summaries.** `src/Display.jl` gives every result type a one-line `show` and a short `text/plain` summary. A new result type gets both, plus an entry in `_DISPLAY_FIELDS`; a new field on a summarized type is classified there as shown or omitted, which `test/display_check.jl` enforces.
 

Project.toml

@@ -25,8 +25,6 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
 OrdinaryDiffEq = "1dea7af3-3e70-54e6-95c3-0bf5283fa5ed"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
-PlotlyJS = "f0f68f2c-4968-5e81-91da-67840de0976a"
-Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 QuadGK = "1fd47b50-473d-5c70-9696-f719f8f3bcdc"
@@ -39,6 +37,12 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 TOML = "fa267f1f-6049-4f14-aa54-33bafae1ed76"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
+[weakdeps]
+Makie = "ee78f7c6-11fb-53f2-987a-cfe4a2b5a57a"
+
+[extensions]
+GPECMakieExt = "Makie"
+
 [compat]
 AdaptiveArrayPools = "0.3.5"
 CommonSolve = "0.2"
@@ -57,10 +61,9 @@ JLD2 = "0.6.3"
 LaTeXStrings = "1.4.0"
 LinearAlgebra = "1"
 Logging = "1"
+Makie = "0.24"
 OrdinaryDiffEq = "6.102.0"
 Pkg = "1"
-PlotlyJS = "0.18.17"
-Plots = "1.40.15"
 PrecompileTools = "1.2"
 Printf = "1"
 QuadGK = "2.11.3"