epkg/tests · openEuler/epkg

Wwu_fengguangtests: add automatic update after env creation in dev-projects tests

文件	最后提交记录	最后更新时间
busybox	tests: remove unnecessary 2>/dev/null \|\| true from busybox-runtest.sh Remove '2>/dev/null \|\| true' from the tar.tempdir cleanup line. Change from: rmdir tar.tempdir 2>/dev/null \|\| true To: [ -d tar.tempdir ] && rmdir tar.tempdir This preserves error context - if rmdir fails for a real reason (permissions, not empty, etc.), we want to see the error. The check ensures we only try to remove if the directory exists. Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前
cross-platform	tests/cross-platform: add --assume-yes to avoid stall Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	2 个月前
dev-projects	tests: add automatic update after env creation in dev-projects tests Problem: When running dev-projects tests on Fedora, package installation fails because the environment's package metadata is stale. The cached repodata references old package versions (e.g., java-25-openjdk-25.0.2.0.10-5.fc43) that no longer exist on mirrors, causing 404 errors during download. Solution: 1. Add run_update() function to common.sh for test scripts that need it 2. Call 'epkg update' automatically in create_env() after environment setup to ensure fresh package metadata before any install operations This ensures all subsequent package installs use current repository metadata. Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	2 个月前
fuzz	tests: improve PIR fuzz test compatibility and error handling Problem/Purpose: - PIR fuzz test failed when .epkg/cache are tmpfs mounts (not symlinks) - Transient network errors counted as failures - tor-gencert shows help but returns exit code 1 (upstream behavior) - fc-cache post-install script failures counted as errors Solution: - Accept tmpfs mount as valid alternative to symlink for .epkg - Accept actual cache directory when CACHE_DIR points to it - Ignore "Network unreachable" errors in whitelist - Ignore "File size mismatch after write" errors during concurrent download - Add tor-gencert to exe_whitelist (shows syntax help, exit code 1) - Add "failed to write cache" to whitelist (fc-cache post-install failure) Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前
in-vm	tests: fix env-register-activate test for idempotent execution Fix two issues causing confusing error output: 1. Remove stale test environments before creation - test-install-env, test-pure-env, test-stack1-env, test-stack2-env - These could remain from previous test runs, causing creation failures 2. Fix 'main' environment public/private test - Old test tried `epkg env create main --public` expecting --public rejection - But actually failed because main already exists - Then used `env config get public` which doesn't support -e flag - New test directly reads main's env.yaml to verify private status Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	2 个月前
in-windows	tests: remove set -e from test scripts per tests/README.md principles Remove 'set -e' from the following test scripts: - busybox/busybox-runtest.sh - busybox/mk-epkg-bindir.sh - busybox/run-one.sh - in-windows/run-tests.sh - lua/test_compare.sh - lua/test_lua.sh Per tests/README.md principles: - Test scripts should not use 'set -e' to preserve context and error info - Debugging failed tests requires seeing the full context, not immediate exit Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前
lua	tests: add open_max to known musl/glibc differences musl returns a static large value (524288) for sysconf(_SC_OPEN_MAX) while glibc queries the actual runtime limit via getrlimit(). Both are POSIX compliant behaviors - this is a design difference. Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	2 个月前
misc	tests: fix color output using printf instead of echo Purpose/Problem: Shell scripts use ANSI escape codes like \033[0;32m for colored output, but plain 'echo' does NOT interpret these sequences - they are printed as literal text. This results in output like \033[0;32m[TEST]\033[0m instead of colored [TEST]. Solution: 1. Use 'printf %b' which reliably interprets escape sequences across all POSIX shells (unlike 'echo -e' which is bash-specific) 2. Auto-detect tty in set_color_names() - disable colors when output is piped/redirected (non-tty context) 3. Provide unified color_print(), log(), error(), skip() functions in common.sh that scripts can reuse or override Changes: - tests/common.sh: add tty detection in set_color_names() - tests/common.sh: add color_print() helper function - all test scripts: replace 'echo "\${GREEN}..."' with 'printf %b' Benefits: - Colors work correctly on terminal - No color codes clutter when output is piped/redirected - POSIX-compliant, works across bash/dash/ash/etc Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前
osroot	expose: fix ebin wrapper symlink resolution for multi-distro support ## Problem The ebin wrappers were generated using store paths directly, which broke for distros that use symlinks differently: - Alpine: /usr/bin/npm -> ../share/nodejs/npm/bin/npm-cli.js - OpenEuler: /usr/bin/npm -> ../share/nodejs/npm/bin/npm (shell script) - Debian: /usr/bin/nodejs -> node, /usr/bin/npm -> ../share/nodejs/npm/bin/npm-cli.js The wrappers pointed to wrong paths because symlink resolution was done in the store context, not the env context. ## Solution 1. Added resolve_ebin_target_path() function that: - Starts from the env path (e.g., $env_root/usr/bin/npm) - Follows symlinks within the env context - Stops when hitting a real file or symlink staying inside env - Returns the resolved path for use in the wrapper 2. Updated get_exec_command() to: - Accept already-resolved env paths (no store-to-env conversion needed) - Handle npm/npx shell scripts specially (call -cli.js directly) 3. Added comprehensive block comments explaining the ebin wrapper design ## Testing Added unit tests in src/expose.rs with real-world symlink layouts from: - Alpine (npm, npx -> npm-cli.js, npx-cli.js) - OpenEuler (npm -> npm shell script) - Debian (nodejs -> node, npm -> npm-cli.js) Test command: cargo test resolve_ebin_target_path ## Verified - ./tests/dev-projects/run.sh -o alpine -t node - PASSED - ./tests/dev-projects/run.sh -o openeuler -t node - PASSED - cargo test resolve_ebin_target_path - 6 tests PASSED	4 个月前
sandbox	tests: fix VM lifecycle tests by specifying timeout=0 Problem: - Test VM-1 and VM-10 failed because VM shutdown immediately after the first command completed - Root Cause: `vm start` without `-s timeout=N` uses default timeout=None - `vm_keep_timeout_secs=None` triggers vm-daemon to return Shutdown disposition, causing immediate VM shutdown after command completes Analysis (from VM log): [ 1.030578] run_vsock_server thread: Shutdown disposition [ 1.109159] run_vsock_server: shutdown requested, exiting [ 1.115629] sysrq: Power Off - VM shutdown at 1.1s after `echo ready` command in wait_for_guest_ready() - Test VM-2/VM-3 passed incorrectly because zombie daemon process was still detected as "alive" by is_process_alive() - Test VM-4 failed connecting to already-shutdown VM Solution: - Add `-s timeout=0` to Test VM-1 and VM-10 `vm start` commands - timeout=0 means "never auto-shutdown", keeping VM alive for reuse Fixes: ./test-vm-sandbox.sh --vmm=qemu Test VM-4 error "Expected 'reuse test', got ' src/vm/client.rs:135'" Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	2 个月前
solver	tests: refactor to use GenerationCommand instead of InstallationPlan Refactored solver test system to use GenerationCommand for test expectations instead of InstallationPlan. This simplifies test file format and removes the need for custom deserialization logic. Changes: - Added plan_to_generation_command() helper function in plan.rs to extract package lists and expose operations from InstallationPlan - Updated TestCaseMetadata to use GenerationCommand instead of InstallationPlan - Removed custom Deserialize implementation for InstallationPlan (no longer needed since tests deserialize GenerationCommand directly) - Removed Serialize derive and serde attributes from InstallationPlan struct - Refactored record_history() to use plan_to_generation_command() helper - Updated validate_plan() to use helper and compare directly with GenerationCommand Vec fields - Converted all test YAML files from map format (pkgkey: null) to Vec format (pkgkey list), making test files cleaner and easier to read - Added serde(default) attributes to GenerationCommand metadata fields for proper deserialization Test file format change: Before: fresh_installs: {pkgkey: null} After: fresh_installs: [pkgkey] This provides better separation of concerns: InstallationPlan is the internal runtime structure, while GenerationCommand is the serialized representation used for history and tests. Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	6 个月前
README.md	tests: add run-tests.sh entry point for Windows native tests in WSL2 Problem: - Windows batch tests (tests/in-windows/.cmd) cannot run directly from WSL2 because cmd.exe does not support UNC paths (\\wsl.localhost\...) - No way to run individual test cases from WSL2 - EPKG_BIN environment variable not supported in run-tests.cmd Purpose: Enable running Windows native tests from WSL2 with single test selection capability, similar to tests/in-vm/test-one.sh interface. Solution: 1. Add run-tests.sh - WSL2 entry point that: - Copies tests and binary to Windows-accessible temp location (C:\temp_epkg_test_) - Runs tests via cmd.exe from C: drive to avoid UNC path issues - Supports EPKG_BIN environment variable for custom binary path - Supports -d/-dd/-ddd debug flags - Supports selecting individual test case: env-path-auto-discovery, export-import, history-restore, install-remove-upgrade (or iur), all 2. Modify run-tests.cmd to: - Accept optional second argument for test name - Add goto labels for running individual tests - Add :done label for proper exit code handling Usage: cd tests/in-windows ./run-tests.sh # Run all tests ./run-tests.sh export-import # Run single test ./run-tests.sh -dd install-remove-upgrade # Debug mode EPKG_BIN=dist/epkg.exe ./run-tests.sh # Custom binary Examples: ./run-tests.sh ./run-tests.sh iur ./run-tests.sh -dd history-restore Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前
common.sh	tests: add has_cmd helper and use it in test scripts Add has_cmd() function to tests/common.sh for checking command existence. This is a cleaner alternative to 'command -v ... >/dev/null 2>&1'. Update test scripts to use has_cmd: - sandbox/test-vm-sandbox.sh: check_dependency(), auto_select_vmm(), virtiofsd checks - in-vm/cases/public-multi-user.sh: adduser/useradd detection - in-vm/cases/env-register-activate.sh: jq command verification Signed-off-by: Wu Fengguang <wfg@mail.ustc.edu.cn>	3 个月前

Tests shall be able to run freely inside AI agent sandbox, so that AI agent can freely reproduce-debug-fix bugs in a fully automated loop.

Test scripts shall follow below common principles:

Supports debug mode with -d/-dd/-ddd flags
do not 'set -e'
avoid lots of >/dev/null: we are testing! so preserve context and error info
Assumes epkg is already installed (except for the more heavier tests/in-vm/ which covers install-from-scratch tests and tests that may pollute host os)
(Re-)creates new env with non-random names for various testing
Run tests with 'timeout' prefix and '-y|--assume-yes' for automation w/o blocking
Log to /tmp/ files with non-random names with backup file for grep based problem analyze and comparison with history behavior
Leaves the env for human/agent debug; i.e. do not remove env in the end, but remove it in the beginning, before create, if it already exists

Test Structure

/c/epkg/tests/
├── README.md               # This file - common test principles
├── common.sh               # Shared shell functions (parse_debug_flags, set_epkg_bin, etc.)
│
├── in-vm/                  # In-VM tests (microVM-based isolation)
│   ├── README.md           # In-VM test documentation
│   ├── cases/              # Test cases (one per feature)
│   ├── host-vars.sh        # Host-side variables
│   ├── vars.sh             # Guest-side variables
│   ├── lib.sh              # Shared shell helpers
│   ├── vm.sh               # VM launcher
│   ├── entry.sh            # Guest entry point
│   ├── test-one.sh         # Run single test
│   ├── test-all.sh         # Run all tests
│   ├── test-iur.sh         # Install/Remove/Upgrade matrix
│   ├── e2e-combo.sh        # VMM/memory knobs wrapper
│   ├── run-alpine-vm.sh    # Alpine-focused tests
│   └── test-dev.sh         # Build-from-source tests
│
├── sandbox/                # Sandbox/isolation tests
│   ├── test-vm-sandbox.sh  # VM sandbox backend tests
│   └── test-isolation-modes.sh  # env/fs/vm isolation tests
│
├── dev-projects/
│   ├── langs/              # Language toolchain tests
│   │   ├── go.sh           # Go toolchain tests
│   │   ├── python.sh       # Python toolchain tests
│   │   ├── rust.sh         # Rust toolchain tests
│   │   └── [other languages]   # java.sh, node.sh, etc.
│   └── scenes/             # Test scenarios (environment behaviors)
│       ├── bash.sh         # Bash installation and /bin/sh tests
│       ├── curl-https.sh   # curl installation and HTTPS tests
│       ├── epkg-nested.sh  # Nested epkg via bash tests
│       ├── package-queries.sh  # rpm/dpkg query tests
│       └── search-paths.sh     # epkg search --paths tests
│
├── misc/                   # Miscellaneous tests
│   └── test-shell-wrapper.sh  # epkg.sh wrapper tests
│
├── solver/                 # Dependency solver unit tests
├── busybox/                # Busybox-specific tests
├── lua/                    # Lua-specific tests
├── osroot/                 # OS root filesystem tests
├── cross-platform/         # Cross-platform tests
└── in-windows/             # Windows native tests (run from WSL2)
    ├── run-tests.cmd       # Windows test runner (native Windows)
    └── run-tests.sh        # WSL2 entry point for running Windows tests

Running Tests

Quick Tests (host-safe, no VM required)

# Scenario tests
cd /c/epkg/tests/dev-projects/scenes
E2E_OS=debian ./bash.sh
E2E_OS=alpine ./curl-https.sh

# Sandbox isolation tests
cd /c/epkg/tests/sandbox
E2E_OS=debian ./test-isolation-modes.sh

# Shell wrapper tests
cd /c/epkg/tests/misc
./test-shell-wrapper.sh

In-VM Tests (VM-based isolation)

cd /c/epkg/tests/in-vm

# Run all tests
./test-all.sh

# Run single test
./test-one.sh cases/env-register-activate.sh
./test-one.sh -d cases/bare-rootfs.sh  # Debug mode

# With specific OS
E2E_OS=alpine ./test-one.sh cases/bare-rootfs.sh

# With specific VMM/Memory
E2E_VMM=qemu E2E_VM_MEMORY=8G ./e2e-combo.sh cases/install-remove-upgrade.sh

# Alpine-focused
./run-alpine-vm.sh

# Install/Remove/Upgrade matrix
./test-iur.sh

# Build-from-source
./test-dev.sh

Windows Native Tests (from WSL2)

Run Windows-native batch tests from WSL2 using the shell entry point:

cd /c/epkg/tests/in-windows
./run-tests.sh                           # Run all tests
./run-tests.sh export-import             # Run single test
./run-tests.sh -dd install-remove-upgrade # Run with debug logging
EPKG_BIN=dist/epkg.exe ./run-tests.sh    # Use specific binary

Available test names: env-path-auto-discovery, export-import, history-restore, install-remove-upgrade (or iur), all

This script copies the tests and binary to a Windows-accessible location (C:\temp_epkg_test_*) and runs them via cmd.exe, because Windows cmd.exe cannot work with WSL UNC paths (\\wsl.localhost\...).

Requirements:

WSL2 with Windows C: drive mounted at /mnt/c
Windows epkg.exe binary (built via make cross-windows)

Debug Flags

All test scripts support consistent debug flags:

Flag	Effect
`-d` / `--debug`	Interactive mode (pause on error)
`-dd`	Debug logging (`RUST_LOG=debug`)
`-ddd`	Trace logging (`RUST_LOG=trace`, `RUST_BACKTRACE=1`)

Common Functions (common.sh)

parse_debug_flags "$@" - Parse -d/-dd/-ddd flags
set_epkg_bin - Find and set EPKG_BIN
set_color_names - Set up color variables

Environment Variables

Variable	Used In	Description
`E2E_OS`	All	Target OS/distro (debian, alpine, fedora, etc.)
`E2E_VMM`	E2E, Sandbox	VMM backend (qemu, libkrun)
`E2E_VM_MEMORY`	E2E	VM RAM (default: 16G)
`E2E_BACKEND`	Guest	Set to `vm` in VM guests
`RUST_LOG`	All	Rust logging level
`RUST_BACKTRACE`	All	Enable backtraces