makepad/libs/makepad_test

makepad_test

makepad_test provides Rust-native UI regression tests for Makepad apps. Tests live next to the package they exercise, run through normal cargo test, and drive an owned standalone release app through its --remote HTTP surface. Test windows are hidden by default; no Studio instance or hub is required.

Quick Start

Add this to the package under test:

[dev-dependencies]
makepad-test = { path = "../../libs/makepad_test", version = "0.1.0" }

Create an integration test:

use makepad_test::{makepad_test, Selector, TestApp};

#[makepad_test]
fn fill_and_submit(app: TestApp) {
    app.locator(Selector::id("input_singleline"))
        .wait_visible()
        .fill("hello")
        .wait_value("hello");
    app.press_return();
    app.locator(Selector::id("status_label"))
        .wait_text("Returned from singleline: \"hello\"");
}

Run a package-local suite with:

cargo test --release -p makepad-example-text-input --test ui -- --test-threads=1

To watch a standalone test window, opt into visible mode. It remains unfocused:

MAKEPAD_TEST_VISIBLE=1 \
MAKEPAD_TEST_STARTUP_DELAY_MS=1000 \
MAKEPAD_TEST_ACTION_DELAY_MS=750 \
MAKEPAD_TEST_KEEP_OPEN_MS=3000 \
cargo test --release -p makepad-example-text-input --test ui -- --test-threads=1

Each test builds from its package directory, reusing Cargo's owning-workspace release target. An inherited CARGO_TARGET_DIR or explicit TestConfig::env["CARGO_TARGET_DIR"] override is respected; the harness does not create a separate build target per example.

The runner starts only its own executable, reads its PID and ephemeral port from the startup line, and closes it with /gq (grab and quit), falling back to /quit when needed. It waits for the owned process to exit and kills that exact child only if graceful shutdown fails. Existing user instances are never reused.

Surface Area

• #[makepad_test] for current-package UI tests
• TestApp for app-scoped input, waits, logs, screenshots, and low-level input forwarding
• Selector for structured snapshot matching
• Locator for strict single-widget interaction and assertions

Structured selectors support:

• Selector::all()
• Selector::id("...")
• Selector::widget_type("...")
• Selector::raw("...")
• builder filters: .text_exact(...), .text_contains(...), .nth(...), .window(...), .window_index(...), .any_window()

Common locator actions:

• click, type_text, fill, clear
• press_key, press_key_with_modifiers
• scroll, drag_by, touch_drag_by (and _and_hold variants that don't fling)
• touch_drag_through_and_hold, a finger that turns through several offsets

Common waits and assertions:

• wait_visible, wait_hidden, wait_count
• wait_text, wait_value, wait_checked, wait_enabled
• assert_text, assert_value, assert_checked, assert_enabled

Inspection helpers:

• widget_snapshot()
• widget_dump()
• screenshot()
• wait_for_log_contains(...)

Failure Artifacts

Artifacts are package-local (separate from the shared Cargo build target):

<manifest_dir>/target/makepad_test/<package>/<test>/

Builds write build-stderr.txt. Launched sessions also write app-stdout.txt, app-stderr.txt, and shutdown.txt. Test-body failures additionally capture:

• failure.txt
• logs.txt
• widget-snapshot.json
• widget-tree.txt or widget-tree-error.txt
• failure-screenshot.png or failure-screenshot-error.txt

Current Constraints

• synchronous API only
• the macro targets its current package; run_with_config can target another manifest/package
• forward supports pointer, scroll, key, and text input; other legacy protocol messages return an error
• widget snapshots require the current remote fields window_id and boolean enabled; optional selected is preserved
• milestone-1 repo suite is validated on macOS first
• no visual diffing or trace viewer yet

Guide

For the full authoring model, runtime behavior, and troubleshooting notes, see GUIDE.md.