test(map): a corpus of QGIS projects and styles the importer reads

ATTRIBUTION.md

@@ -12,6 +12,7 @@ which every package ships next to LICENSE. They are generated from `Cargo.lock`
 |---|---|---|
 | Sample data (`countries`, `countries_chn`, `cities`, `rivers`, `lakes` GeoJSON), embedded at build time from [rustgis-assets](https://github.com/opengeos/rustgis-assets/tree/main/data) | Natural Earth 1:110m, attributes trimmed; `countries_chn` is Natural Earth's China point of view (1:10m, simplified) | Public domain |
 | QGIS screenshot (`qgis-world-atlas.webp` in [rustgis-assets](https://github.com/opengeos/rustgis-assets/tree/main/images)), on the [RustGIS and QGIS](docs/comparison/qgis.md) page | Taken by the RustGIS project: [QGIS](https://qgis.org) 4.2 showing the Natural Earth sample data above | QGIS is GPL-2.0-or-later; the data is public domain |
+| QGIS test corpus (`crates/map/testdata/qgis/generated/`: styles, projects and their sample data) | Dominik Strebel ([@pick2510](https://github.com/pick2510)) for the RustGIS project: written by [QGIS](https://qgis.org) 4.2.3 through its API, from `generate.py` there; the data is made up | Apache-2.0 |
 
 Raster coordinate systems are looked up by EPSG code in the PROJ definitions of the
 [crs-definitions](https://crates.io/crates/crs-definitions) crate (CC0-1.0), compiled into the

CITATION.cff

@@ -17,6 +17,8 @@ authors:
     family-names: Hofbauer
   - given-names: Mavrick
     family-names: Fitzgerald
+  - given-names: Dominik
+    family-names: Strebel
 version: 0.4.0
 date-released: 2026-10-25
 doi: "10.5281/zenodo.23633281"

crates/map/src/qgis/tests.rs

@@ -535,3 +535,40 @@ fn raster_renderers_read_as_rustgis_raster_styles() {
         assert!(s.raster.is_none() && report.notes.iter().any(|n| n.text.contains("couldn't read")), "{bad}: {report:?}");
     }
 }
+
+/// Every generated style and project in `testdata/qgis` reads, each project with its layers.
+#[test]
+fn every_corpus_file_reads() {
+    let root = std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("testdata/qgis");
+    let files = |dir: &str, extensions: &[&str]| {
+        let mut paths: Vec<_> = std::fs::read_dir(root.join(dir)).unwrap().map(|e| e.unwrap().path()).collect();
+        paths.retain(|path| path.is_file() && path.extension().and_then(|e| e.to_str()).is_some_and(|e| extensions.contains(&e)));
+        paths.sort();
+        paths
+    };
+    // Pin the corpus sizes so a regeneration cannot silently drop fixtures.
+    let styles = files("generated/qml", &["qml"]);
+    assert_eq!(styles.len(), 90, "style corpus changed: {styles:?}");
+    for path in &styles {
+        let text = std::fs::read_to_string(path).unwrap();
+        let mut report = Report::default();
+        read_style(&text, "layer", &mut report).unwrap_or_else(|e| panic!("{}: {e}", path.display()));
+    }
+    let projects = files("generated/qgs", &["qgs", "qgz"]);
+    assert_eq!(projects.len(), 21, "project corpus changed: {projects:?}");
+    for path in &projects {
+        let bytes = std::fs::read(path).unwrap();
+        let text = if path.extension().is_some_and(|e| e == "qgz") {
+            let entries = rustgis_data::zip::read_matching_within(&bytes, |n| n.ends_with(".qgs"), 64 << 20).unwrap();
+            String::from_utf8(entries.into_iter().next().unwrap().1).unwrap()
+        } else {
+            String::from_utf8(bytes).unwrap()
+        };
+        let dir = path.parent().unwrap().to_string_lossy().to_string();
+        let mut report = Report::default();
+        let project = read_project(&text, &dir, &mut report).unwrap_or_else(|e| panic!("{}: {e}", path.display()));
+        assert!(!project.layers.is_empty(), "{}: no layers", path.display());
+        // The PostGIS and SQL Server projects' made-up password reaches neither the layers nor the notes.
+        assert!(!format!("{project:?} {report:?}").contains("secret"), "{}: the password came through", path.display());
+    }
+}

crates/map/testdata/qgis/README.md

@@ -0,0 +1,51 @@
+# QGIS file corpus
+
+By Dominik Strebel ([@pick2510](https://github.com/pick2510)), for the RustGIS project.
+
+QGIS projects (`.qgs`, `.qgz`) and layer styles (`.qml`) for testing the QGIS importer
+(`crates/map/src/qgis/`, the `project.importQgis` and `style.importQgis` commands). QGIS publishes
+no specification of these formats, so the corpus stands in for one: `every_corpus_file_reads` in
+`src/qgis/tests.rs` reads each file, and the files show what real projects contain.
+
+| Folder | What | Licence |
+|---|---|---|
+| `generated/qml/` | 90 styles written by QGIS 4.2.3 through its API, each varying one thing: every symbol layer type with its defaults, every renderer (categorized and graduated by expression, rule-based with nested and else rules), every paint effect, data-defined size, angle and colour, labels (every placement, buffer, background, shadow, font, wrap, expression text, offsets, data-defined settings, scale ranges, curved labels, each callout, rule-based labels), layer settings (opacity, blend mode, scale range, reference scale, subset) and raster renderers (grey, pseudocolour interpolated, discrete and exact, multiband, paletted, hillshade, contour) | Apache-2.0, like RustGIS: made by Dominik Strebel ([@pick2510](https://github.com/pick2510)) for the RustGIS project |
+| `generated/qgs/` | 21 projects: one per provider (OGR, GDAL, memory, delimited text by x/y and WKT, SpatiaLite, virtual, WMS, XYZ, WFS, WCS, ArcGIS feature and map server, vector tiles, PostGIS, SQL Server, GPX); project structure (groups, a mutually exclusive group, a map theme, a bookmark, a variable, a relation, a join, a virtual field, time from a field, two styles on a layer); a layout with every item and an atlas; the annotation layer with every item; `all-styles.qgz`, every style above on its own layer | Apache-2.0 |
+| `generated/data/` | The data those files use: `sample.gpkg` (layers `points`, `lines` and `polygons`, 9 made-up features each, fields `id`, `name`, `kind`, `value`, `angle`, `day`), `dem.tif` (3 bands of a linear surface), two CSVs, a GPX file | Apache-2.0 |
+| `generated/generate.py`, `scrub.py` | The generator, and the step that removes the user's name and absolute paths from what QGIS writes | Apache-2.0 |
+| `generated/types.txt` | The element types QGIS 4.2.3 knows (symbol layers, renderers, providers, label placements, callouts, annotation and layout items, paint effects): the checklist the corpus was made from. A few have no file yet: the single-colour raster renderer, and the text table, fixed table, elevation profile and chart layout items | Apache-2.0 |
+
+## Not included
+
+- **The SMV25 project and its data.** The repository does not bundle third-party vendor styles.
+  For a large real-world project to test manually, download the GeoPackage version and its
+  QGIS project from
+  [swisstopo](https://www.swisstopo.admin.ch/en/national-map-swiss-map-vector-25) (also on
+  [opendata.swiss](https://opendata.swiss/en/dataset/swiss-map-vector-25-vektorielle-landeskarte-der-schweiz-1-25-000))
+  and place the project beside `SMV25_CHLV95LN02.gpkg` and the four rasters in
+  `SMV25_CHLV95LN02_RASTER/` (about 760 MB) to open the whole map. These external files are
+  outside the automated corpus test.
+- **The QGIS Training Manual's projects and styles**, which were part of the audit: they are
+  GPL-2.0, so they live in [qgis/QGIS-Training-Data](https://github.com/qgis/QGIS-Training-Data)
+  (`exercise_data/`, 36 `.qgs`, `.qgz` and `.qml` files from QGIS 1.8 to 3.x).
+- **`sample.sqlite`**, the SpatiaLite copy of `sample.gpkg` the SpatiaLite project reads (7 MB,
+  mostly SpatiaLite's table of coordinate systems): `generate.py` writes it.
+- The service and database layers (WFS, WCS, ArcGIS, PostGIS, SQL Server) point at
+  `example.invalid` or `db.invalid`; QGIS writes them as configured, without data. The PostGIS
+  and SQL Server sources carry a made-up password (`secret`), to test that the importer drops it.
+
+## Regenerating
+
+QGIS 4.2.3 (here the Flathub Flatpak, whose `/tmp` is private, so it writes under `~/.cache`):
+
+```sh
+mkdir -p ~/.cache/qgis-corpus
+flatpak run --env=PYTHONPATH=/app/share/qgis/python --env=QT_QPA_PLATFORM=offscreen \
+  --command=python3 org.qgis.qgis generated/generate.py ~/.cache/qgis-corpus/out
+```
+
+It writes `data/`, `qml/`, `qgs/`, `types.txt` and `log.txt` (what was written, and what this
+QGIS couldn't write) into the output folder, then runs `scrub.py` on the projects. QGIS may crash
+on exit, after writing everything. Copy the files here without `log.txt`, `sample.sqlite`,
+`dem.tif.aux.xml`, `all-styles.qgs` (the `.qgz` holds it) and the `_attachments.zip` files QGIS
+writes beside each project.

crates/map/testdata/qgis/generated/data/points-wkt.csv

@@ -0,0 +1,3 @@
+name,wkt
+A,POINT(2697000 1261800)
+B,"LINESTRING(2697000 1261800, 2697100 1261900)"

crates/map/testdata/qgis/generated/data/points-xy.csv

@@ -0,0 +1,3 @@
+name,x,y
+A,2697000,1261800
+B,2697100,1261900

crates/map/testdata/qgis/generated/data/waypoints.gpx

@@ -0,0 +1 @@
+<?xml version="1.0"?><gpx version="1.1" creator="corpus" xmlns="http://www.topografix.com/GPX/1/1"><wpt lat="47.5" lon="8.73"><name>A</name></wpt></gpx>