Files
blog-post-backup/designs/grel-rs/TECHNICAL_DESIGN.md

13 KiB
Raw Blame History

📘 grel-rs Technical Design Document (v3.0)

Binary: grel | Repository: grel-rs
Target Platforms: Linux, Windows (macOS optional)
Core Philosophy: Pure CLI, deterministic asset resolution, pacman/Artix-familiar UX, explicit over implicit, robust upgrade edge-case handling.


1. Overview & Goals

grel (Global/General Release) is a terminal-native, high-performance package manager for downloading and managing binary releases from Git forges. It abstracts provider-specific APIs into a unified resolution pipeline, supports HTTP/SOCKS5 proxies, caches optimal DNS endpoints, downloads assets in parallel, and delivers a transparent, scriptable, pacman-compatible UX.

Key Requirements Met:

  • Pure CLI (no TUI/GUI), standard std::io prompts
  • Deterministic asset resolution: filter → sort → explicit prompt (zero scoring magic)
  • Robust parsing of non-standard release names via Os/Arch enums + alias mapping
  • Format control: .deb/.rpm/.msi/.dmg disabled by default
  • Pacman/Artix-style numbered selection with [default], TTY/CI fallback, -y/--noconfirm
  • -Syu resilience: detects filename renames, handles repo 404s via orphan tracking, explicit migration path
  • Cross-platform PATH integration (grel path add), XDG-compliant, no sudo
  • Forge-agnostic provider routing (public + self-hosted URLs)

2. Workspace Architecture

grel-rs/
├── Cargo.toml                  # Workspace root (resolver = "2")
├── crates/
│   ├── grel-cli/               # CLI parsing, pure-text prompts, progress routing
│   ├── grel-core/              # Resolution, tokenization, filter/sort pipeline, upgrade state
│   ├── grel-providers/         # Forge trait, registry, API implementations (modules)
│   ├── grel-network/           # HTTP client, proxy routing, IP cache resolver, parallel downloader
│   ├── grel-cache/             # SQLite state, IP cache, artifact storage, TTL eviction
│   └── grel-config/            # Layered config, migrations, asset priority matrices
├── tests/                      # Integration, mock servers, e2e fixtures
└── scripts/                    # CI, release, benchmark helpers
Crate Responsibility
grel-cli Subcommand routing, pure CLI prompt loops, indicatif progress, tabled output
grel-core PackageRef parsing, AssetTokens extraction, deterministic filter/sort, upgrade planning
grel-providers ReleaseProvider trait, GitHub/GitLab/Gitea/Codeberg modules, self-hosted auto-detection
grel-network Proxy chaining, DNS/IP cache resolver, JoinSet parallel downloads, streaming extraction
grel-cache state.sqlite management, ETag/IP cache, LRU artifact eviction, atomic install temp dirs
grel-config TOML loading, env/CLI overrides, schema validation, config migrations

3. CLI Specification & Pure UX

No TUI frameworks. All interaction uses standard terminal I/O with predictable, pipe-safe behavior.

Commands & Aliases

Command Aliases Description
grel sync foo/bar S, -S Download & install latest
grel sync foo/bar@1.2.3 -S Pin version
grel search keyword Ss, -Ss Search across registered forges
grel upgrade Syu, -Syu Refresh + upgrade installed
grel info foo/bar Si, -Si Show release metadata
grel list Q, -Q List installed packages (shows status)
grel remove foo/bar R, -R Uninstall & clean DB
grel clean Sc, -Sc Purge artifact/metadata cache
grel path add - Generate shell/registry snippets for $PATH
grel migrate old/path new/path - Remap repo path in state DB

Pure CLI Prompt (Artix/Pacman Compatible)

:: 3 compatible asset(s) found for linux/x86_64
  1) foo-1.0.0-linux-x86_64.tar.gz | 12.4 MB | tar.gz (default)
  2) foo-1.0.0-linux-x86_64-musl.tar.gz | 10.2 MB | tar.gz
  3) foo-1.0.0-linux-amd64.tar.gz | 11.8 MB | tar.gz
:: Select asset to download [1]: _
  • Accepts 13 or Enter (selects [1])
  • Invalid input → :: Invalid selection. Enter a number [1-3]:
  • --noconfirm / -y → skips prompt, selects [1]
  • Non-Interactive Fallback: !std::io::stdin().is_terminal() → auto-selects [1], prints Non-interactive mode. Using: <filename> to stderr, never blocks CI/pipes.

4. Asset Resolution Pipeline (Deterministic)

No scoring. No fuzzy logic. Strict filtering → transparent sorting → explicit selection.

Step 1: Tokenization

Filenames are split on non-alphanumeric characters. Tokens are matched against case-insensitive alias maps for Os and Arch.

// foo-v1.2.3-win64-setup.exe → Os::Windows, Arch::X86_64
// bar_1.0.0_linux_amd64.tar.gz → Os::Linux, Arch::X86_64
// app.Darwin.arm64.zip → Os::MacOS, Arch::Aarch64

Unknown tokens fall back to Os::Unknown("...") or Arch::Unknown("...") → never crash, just filter out later.

Step 2: Strict Filtering

Assets are filtered in fixed order. Failure at any step → discard.

1. OS matches target OR is Unknown
2. Arch matches priority list OR fallback allowed (32-bit on 64-bit)
3. Format NOT in `ignore_formats` (deb/rpm/msi/dmg disabled by default)

Step 3: Deterministic Sorting

Remaining assets sorted by explicit cascade:

  1. arch_priority index (lower = better)
  2. prefer_formats index
  3. Lexicographic filename
  4. Size descending (tie-breaker)

Step 4: Selection

  • 0 assets → ❌ No compatible assets. Run grel sync --list-assets
  • 1 asset → Auto-select, log Selected: <filename>
  • >1 assets → Pure CLI numbered prompt (see §3)

5. Platform Enums & Alias Mapping

Strongly-typed, exhaustive, infallible parsing.

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
#[non_exhaustive]
pub enum Os { Linux, Windows, MacOS, FreeBSD, Android, iOS, Unknown(String) }

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
#[non_exhaustive]
pub enum Arch { X86_64, Aarch64, I686, ArmV7, ArmV6, Riscv64, S390x, PowerPC64, Unknown(String) }
  • FromStr implemented with to_lowercase() + alias table
  • serde ready for TOML config
  • Display outputs canonical names for DB/CLI consistency

6. Configuration vs State Management

Strict separation. Humans edit TOML. Machines manage SQLite.

~/.config/grel/config.toml

[general]
version = 1
max_concurrent = 4
proxy = "http://127.0.0.1:7890"

[assets]
ignore_formats = ["*.deb", "*.rpm", "*.msi", "*.dmg", "*.pkg", "*.AppImage"]
prefer_formats = ["*.tar.gz", "*.tar.xz", "*.zip", "*.exe"]
arch_priority = ["x86_64", "aarch64", "x86", "armv7"]
fallback_to_32bit = true
prefer_musl = false

[migrations]
"legacy/old-tool" = "new-org/old-tool"

[upgrade]
check_interval_hours = 6
max_parallel_checks = 10

state.sqlite Schema (installed table)

CREATE TABLE installed (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    forge TEXT NOT NULL,              -- 'github', 'gitlab', or self-hosted base URL
    owner TEXT NOT NULL,
    repo TEXT NOT NULL,
    version TEXT NOT NULL,            -- Strict semver
    asset_filename TEXT NOT NULL,     -- Exact remote filename used
    checksum TEXT,
    install_path TEXT NOT NULL,
    status TEXT NOT NULL DEFAULT 'active' CHECK (status IN ('active', 'orphaned', 'migrated')),
    orphaned_at INTEGER,              -- Unix timestamp when marked unreachable
    last_checked INTEGER,             -- For smart polling intervals
    installed_at INTEGER DEFAULT (strftime('%s', 'now'))
);
CREATE UNIQUE INDEX idx_pkg_unique ON installed(forge, owner, repo);
CREATE INDEX idx_status ON installed(status);

7. -Syu Upgrade Logic & Edge Cases

7.1 Filename Change Detection

Authors often rename assets (x86_64amd64, .tar.gz.tar.xz).

  1. Load stored asset_filename from DB
  2. Resolve new release → filter → sort → get new_best
  3. If new_best.filename != stored.filename:
    ⚠️ Remote asset renamed: old-name.tar.gz → new-name.tar.xz
     Proceeding with update...
    
  4. Download → verify → extract → update DB record with new filename. Proceeds safely.

7.2 Repository Rename / 404 Handling

  1. Provider fetch → 404 or unreachable
  2. Mark status = 'orphaned', set orphaned_at = now()
  3. Skip in future -Syu runs
  4. Warn user:
    ⚠️ Package unreachable: foo/bar (HTTP 404)
     Skipped. Run `grel migrate foo/bar new/path` or `grel remove foo/bar`
    
  5. No auto-migration. Explicit user action required.

7.3 Orphaned Visibility

$ grel list
:: Installed packages (3 active, 1 orphaned)
  🟢 bar/fuzz          1.2.3  linux/x86_64  tar.gz
  🟢 foo/tool          0.9.1  windows/amd64 exe
  🟡 legacy/old-proj   2.0.0  linux/x86_64  tar.gz  [orphaned since 2024-05-01]

8. Networking, Proxy & DNS/IP Caching

  • Proxy Priority: --proxy > GREL_PROXY env > http_proxy/all_proxy > config.toml
  • DNS/IP Cache:
    1. Resolve A/AAAA via hickory-resolver
    2. Parallel TcpStream::connect probes on 443
    3. Store fastest IP + RTT in SQLite with 300s TTL
    4. reqwest::ClientBuilder::resolve(host, ip) forces routing
  • Background Refresh: Idle task pre-warms hot domains, evicts stale entries

9. Rate Limiting & ETag Optimization

GitHub: 60/hr unauth, 5000/hr auth. grel handles gracefully:

  1. ETag Caching: If-None-Match304 Not Modified free. Cached in DB.
  2. Smart Polling: Only checks packages where last_checked < now() - check_interval_hours
  3. Rate Limit Parsing: Reads X-RateLimit-Remaining/Reset. If < 50 → sleep/queue.
  4. Auth Prompt: First run suggests GREL_GITHUB_TOKEN for 5000/hr.
  5. --no-api Fallback: Skips remote checks, uses local timestamps only.

10. PATH Integration & Post-Install

  • Installs to ~/.local/share/grel/bin/ (Linux) / %LOCALAPPDATA%\grel\bin\ (Windows)
  • grel path add prints exact shell/registry snippets:
    export PATH="$HOME/.local/share/grel/bin:$PATH"
    
    [Environment]::SetEnvironmentVariable("Path", "$env:LOCALAPPDATA\grel\bin;$env:Path", "User")
    
  • First run: 💡 Run: grel path add to enable global command access
  • XDG-compliant, no sudo, no system conflicts.

11. Security & Reliability

  • TLS: rustls only (no native OpenSSL)
  • Checksums: sha256 verification before extraction
  • Archive safety: Reject .., absolute paths, external symlinks
  • Atomic installs: Temp dir → verify → rename → DB update
  • No auto-exec: Binaries installed but not run unless invoked
  • Graceful degradation: Network errors → retry/backoff, rate limits → sleep/queue, missing assets → clear error + suggestions

12. Dependency Matrix

# Core & Async
tokio = { version = "1", features = ["full"] }
futures = "0.3"
async-trait = "0.1"

# CLI & UX (Pure CLI, no TUI)
clap = { version = "4", features = ["derive", "wrap_help", "string"] }
indicatif = { version = "0.17", features = ["tokio"] }
tracing-indicatif = "0.3"
tabled = "0.16"
owo-colors = "4"

# Network & Proxy
reqwest = { version = "0.12", features = ["rustls-tls", "json", "socks", "stream"] }
hickory-resolver = "0.24"
dashmap = "6"

# Cache & Storage
sqlx = { version = "0.8", features = ["runtime-tokio-rustls", "sqlite"] }
sha2 = "0.10"
directories = "5"
tar = "0.4"
zip = "2"
zstd = "0.13"
xz2 = "0.1"
bzip2 = "0.5"
sanitize-filename = "0.5"

# Config & Utils
figment = { version = "0.10", features = ["toml", "env"] }
semver = "1"
url = "2"
chrono = { version = "0.4", features = ["serde"] }
regex = "1"
serde = { version = "1", features = ["derive"] }
anyhow = "1"
thiserror = "2"
tracing = "0.1"
tracing-subscriber = { version = "0.3", features = ["env-filter", "fmt"] }

13. Testing & CI Strategy

  • Unit: Token extraction from 20+ messy filenames, filter/sort determinism, config parsing, migration logic
  • Mocked Providers: wiremock for 200/304/403/429/500, rate limit header injection
  • Integration: Download fixtures → extract → verify checksums & paths, grel migrate DB state changes
  • Non-Interactive: echo "" | grel sync foo/bar must not hang, must log to stderr
  • Cross-Platform CI: ubuntu-latest, windows-latest (GitHub Actions)
  • Performance: criterion for DNS/IP cache + parallel download throughput
  • Fuzzing: cargo-fuzz on AssetTokens::from_filename() + archive path validation