andy/mapsize
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 4 Wiki Activity

Docs: architecture, development and performance for the Rust implementation

Browse Source
This commit is contained in:
Andy Dixon
2026-10-08 19:56:43 +01:00
parent 01785850bf
commit c3ed83250b
5 changed files with 209 additions and 263 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ARCHITECTURE.md
+81 -64
View File
@@ -5,36 +5,40 @@ part relies on. It is kept in sync with the code; if they disagree, the code
is wrong or this file is stale — fix one of them.
```
cmd/mapsize CLI entry point, flag parsing, mode selection
cmd/treegen synthetic test-tree generator
internal/brand product name / file extensions (rename in one place)
internal/inventory flat node store, aggregation, indexes, categories
internal/scan bounded concurrent scanner (controller + workers)
internal/platform OS-specific metadata (build-tagged files)
internal/treemap squarified layout, LOD, spatial navigation (pure geometry)
internal/filter query lexer / parser / evaluator
internal/snapshot versioned binary snapshot format + comparison
internal/duplicate staged duplicate detection
internal/export JSON / CSV / text reports
internal/config settings, themes, JSON config file
internal/tui Bubble Tea model, canvas renderer, views, modals
internal/textutil display sanitising, width-aware truncation, units
src/main.rs CLI entry point, flag parsing
src/app.rs mode selection: TUI, reports, snapshots, comparison
src/brand.rs product name / file extensions (rename in one place)
src/inventory/ flat node store, aggregation, indexes, categories
src/scan.rs bounded concurrent scanner (controller + workers)
src/cancel.rs cancellation token shared by scans and background work
src/platform.rs OS-specific metadata and file operations (cfg blocks)
src/treemap/ squarified layout, LOD, spatial navigation (pure geometry)
src/filter/ query lexer / parser / evaluator
src/snapshot/ versioned binary snapshot format + comparison
src/duplicate.rs staged duplicate detection
src/export.rs JSON / CSV / text reports
src/config.rs settings, themes, JSON config file
src/tui/ event loop, model, canvas renderer, views, modals
src/textutil.rs display sanitising, width-aware truncation, units
examples/treegen synthetic test-tree generator
examples/bench in-process benchmarks
```
Dependency direction is strictly downward: `tui` → (`treemap`, `filter`,
`inventory`, `scan`, …); `scan` → (`inventory`, `platform`); `treemap` depends
on nothing but the standard library. The TUI framework (Bubble Tea v2) is only
imported by `internal/tui` and `cmd/mapsize`.
on nothing but the standard library. The terminal library (crossterm) is only
used by `tui`.
## Inventory representation
`inventory.Tree` is an append-only arena of `Node` records addressed by
`NodeID` (`uint32`). Nodes are stored in fixed-size chunks
(`[][]Node`, 65 536 nodes per chunk), so growing the arena never copies or
`NodeId` (`u32`). Nodes are stored in fixed-size chunks
(`Vec<Vec<Node>>`, 65 536 nodes per chunk), so growing the arena never copies or
moves existing nodes and peak memory never doubles during growth.
A node holds: parent ID, first-child / next-sibling IDs (an intrusive child
list — no per-directory slices), name (the single path component, not the full
list — no per-directory vectors), name (the single path component as raw
bytes, not the full
path), kind, flags, own logical size, own allocated size, mtime, mode,
uid/gid, link count, and aggregates (total logical, total allocated, file
count, directory count, error count). Directory-only data — per-category byte
@@ -42,7 +46,7 @@ totals used for colouring — lives in a side table indexed by a directory index
so files do not pay for it.
Full paths are never stored; `Tree.Path(id)` walks parents. Extensions are
interned into a small table and stored as a `uint16`. All sizes are `int64`.
interned into a small table and stored as a `u16`. All sizes are `i64`.
Node IDs are stable for the lifetime of a `Tree`: nodes are never removed or
reordered. A rescan produces a new `Tree`. This is what makes selection
@@ -51,11 +55,11 @@ persistence trivial (see below).
## Scanner worker architecture
```
+------------------ controller goroutine ------------------+
+------------------ controller thread ---------------------+
| owns: Tree (write lock), pending FIFO of dir NodeIDs, |
| hard-link table, visited-dir table |
+----------------------------------------------------------+
| jobs (bounded chan, cap = workers*2) ^ results (bounded chan)
| jobs (bounded channel, cap = workers*2) ^ results (bounded channel)
v |
worker 1 .. worker N (fixed pool, N from --workers / mode)
```
@@ -77,9 +81,9 @@ while the only consumer of that queue is also blocked. mapsize avoids it by
construction:
* Only the controller produces jobs. Workers only produce *results*.
* The controller's main loop is a single `select` that simultaneously offers
the next pending job to the bounded `jobs` channel *and* receives from the
bounded `results` channel. It never performs a blocking send without also
* The controller's main loop is a single `crossbeam_channel::Select` that
simultaneously offers the next pending job to the bounded `jobs` channel
*and* receives from the bounded `results` channel. It never performs a blocking send without also
being ready to receive. Therefore a worker blocked on `results <-` is always
eventually served, and the controller is never blocked on `jobs <-` alone.
* Discovered-but-unscheduled directories wait in the controller's pending FIFO
@@ -94,11 +98,12 @@ directory with 5 M entries neither balloons worker memory nor stalls the UI.
Completion: the controller tracks `inflight` (dispatched directories whose
final chunk has not arrived). The scan is finished when `pending` is empty and
`inflight == 0`. Cancellation (`context.Context`) stops dispatch immediately;
workers check the context between chunks and between entries, and the
controller drains outstanding results so no goroutine leaks.
`inflight == 0`. Cancellation (`cancel::Cancel`, which the controller also
selects on) stops dispatch immediately; workers check the token between
chunks, and the controller drains outstanding results and joins every
worker, so no thread leaks.
Stress tests (`internal/scan/scan_test.go`) run with 1 worker and job/result
Stress tests (`scan::tests`) run with 1 worker and job/result
channel capacity 1 over trees with thousands of directories to prove progress
under maximum backpressure.
@@ -112,26 +117,27 @@ for a 1 M-file tree at depth 20 that is ~20 K ancestor updates instead of
~20 M. There is no reconciliation pass: totals are exact at every moment for
the data seen so far.
The UI reads the tree under `Tree.RLock`; the controller applies batches of
results under `Tree.Lock`, holding it for at most a few milliseconds (it
The tree is shared as `Arc<RwLock<Tree>>`. The UI reads it under the read
lock; the controller applies batches of results under the write lock, holding it for at most a few milliseconds (it
releases the lock between batches). This is the only lock in the data path.
**Concurrency invariants**
1. Only the scan controller (or a snapshot loader before publication) mutates
a `Tree`.
2. Every read of `Tree` node data from another goroutine holds `RLock`.
2. Every read of `Tree` node data from another thread holds the read lock.
3. Nodes are never deleted; a `NodeID` once observed stays valid.
4. Workers share nothing mutable except the bounded channels and atomic
progress counters.
5. No goroutine ever takes `RLock` while already holding it. Go's
`RWMutex` blocks new readers once a writer waits, so a recursive read
lock can deadlock against the scan controller. The TUI takes the lock
once per entry point (`View`, `handleKey`, mouse handlers); helpers they
call assume it is held. Background commands (filter, duplicates,
snapshot save) take their own lock on their own goroutine.
6. After a scan finishes, the UI goroutine may become the writer (move to
trash updates aggregates under `Lock`). Trash is refused while scanning.
5. No thread ever takes the read lock while already holding it. The
standard `RwLock` may block new readers once a writer waits, so a
recursive read lock can deadlock against the scan controller. The TUI
takes the lock once per entry point (render, key and mouse handlers) and
passes `&Tree` to the helpers it calls. Background tasks (filter,
duplicates, snapshot save) take their own lock on their own thread.
6. After a scan finishes, the UI thread may become the writer (move to
trash updates aggregates under the write lock). Trash is refused while
scanning.
## Hard links and sizes
@@ -166,31 +172,41 @@ releases the lock between batches). This is the only lock in the data path.
## TUI event flow
```
scan controller ──(applies under Lock)──> Tree
scan controller ──(applies under write lock)──> Tree
│ atomic progress counters
▼
tea.Tick (10 Hz while scanning) ──> tickMsg ──> Update marks frame dirty
key/mouse/resize ──> Update ──> state change ──> frame dirty
View(): if dirty → RLock tree → layout (if layout dirty) → paint canvas → string
tick (10 Hz while scanning) ─────┐
crossterm events (input thread) ─┼─> select ─> Model::update ─> state change ─> frame dirty
background task results ─────────┘
render: if dirty → read-lock tree → layout (if layout dirty) → paint canvas → write changed lines
```
There is never one message per file: the scanner does not talk to the UI at
all except for a single `scanDoneMsg`. The UI polls at a fixed rate.
`tui::run` puts the terminal in raw mode on the alternate screen and selects
over three crossbeam channels: terminal events from a dedicated input thread,
a timer tick, and messages from background tasks (filter, duplicates,
snapshot save, scan done). There is never one message per file: the scanner
does not talk to the UI at all except for the scan-done notification. The UI
polls its progress counters at a fixed rate.
`View()` is called by Bubble Tea after *every* `Update`. The model therefore
caches the rendered frame and only repaints when something visible changed.
The model caches the rendered frame and only repaints when something
visible changed; several updates in a row cost at most one render.
Expensive per-directory work (sorting children by size) is cached per data
version. During a scan the version advances at most every 400 ms, so the
layout does not jitter at frame rate even though labels and counters update
at the 10 Hz refresh.
A panic hook restores the terminal (leaves the alternate screen, disables
raw mode and mouse reporting) before the panic is reported; SIGTERM quits
cleanly.
## Rendering: the canvas
All drawing goes to an in-memory `Canvas` of cells (grapheme, fg, bg, attrs).
Views and modals paint into the canvas; the canvas is serialised to a string
with minimal SGR changes. Bubble Tea's renderer diffs frames and downsamples
true-colour to the detected profile (256 / 16 / none), so the drawing code
Views and modals paint into the canvas; the canvas is serialised with minimal
SGR changes, converted to the detected colour profile (truecolour / 256 / 16 /
none), and only the changed span of each changed line is written, inside a
synchronized-output block where the terminal supports it. The drawing code
uses RGB throughout. For 256-colour terminals the canvas quantizes colours
itself (redmean distance over the 6×6×6 cube and grey ramp, never the
palette-dependent first 16 entries): the stock conversion maps dark tints to
@@ -226,7 +242,7 @@ visible. Groups are selectable and inspectable.
## Spatial navigation algorithm
`treemap.Neighbour(rects, selected, dir)` is pure geometry:
`treemap::neighbour(blocks, selected, dir)` is pure geometry:
1. Candidates must lie in the requested half-plane: for RIGHT, `c.X ≥
sel.X + sel.W − tolerance` (with the candidate's centre to the right of the
@@ -241,18 +257,18 @@ visible. Groups are selectable and inspectable.
## Terminal resize architecture
`tea.WindowSizeMsg` only records the new size and marks layout dirty. The next
`View()` recomputes panel geometry for the new size class (large / medium /
A resize event only records the new size and marks layout dirty. The next
render recomputes panel geometry for the new size class (large / medium /
small / too-small), reruns the treemap layout for the new treemap rectangle
(never stretches old coordinates), and repaints. Because work happens only in
`View()` and the frame cache absorbs repeated calls, a storm of resize events
(never stretches old coordinates), and repaints. Because work happens only
at render time and the frame cache absorbs repeated calls, a storm of resize events
costs at most one layout per rendered frame. Modal geometry is derived from
the current size every frame, clamped to the screen, and becomes scrollable
when content does not fit.
## Selection persistence
Selection is a `NodeID` (or, for the synthetic LOD group, `-1 - parentID`).
Selection is a `NodeId` (or, for the synthetic LOD group, `-1 - parent`).
Until the user moves the selection it follows the largest block, which
changes while a scan runs; after that it is sticky. After any relayout the renderer looks the ID up in the
new rectangle list. If it is absent (filtered out, collapsed into a group,
@@ -261,8 +277,8 @@ nearest visible ancestor, then the largest block.
## Modal architecture
A modal is a small interface: `Title()`, `Lines(width)`, `Footer()`, and
`HandleKey`. A generic frame renders any modal centred, clamped to the screen
A modal is a small trait: `title()`, `content(width)`, `footer()`, and
`key`. A generic frame renders any modal centred, clamped to the screen
(min 1-cell margin), with a scrollable viewport when content exceeds height.
The base screen is dimmed before the modal is painted. While a modal is open
it receives all keys; `Esc` always closes it.
@@ -286,7 +302,7 @@ No other package switches on `runtime.GOOS`.
Filenames and snapshot contents are untrusted.
* Every string that reaches the terminal passes through
`textutil.Sanitize`, which replaces C0/C1 control characters (including
`textutil::sanitize`, which replaces C0/C1 control characters (including
ESC, CSI, BEL, DEL) with visible escapes (`\x1b`), replaces invalid UTF-8
with U+FFFD, and neutralises bidi override characters. The canvas also
refuses control characters as a second line of defence. Internal names keep
@@ -299,9 +315,10 @@ Filenames and snapshot contents are untrusted.
(≤ 1 EiB) and node count (at most one node per byte of snapshot file, 1 Mi floor); caps the decompressed body at 64× the file size
(decompression bombs fail fast); and verifies a SHA-256 trailer.
Aggregates are recomputed with saturating arithmetic rather than trusted.
* No shell is ever invoked. "Open/reveal" and trash use direct `exec` with
argument vectors or native APIs; paths are passed as single arguments.
* No shell is ever invoked. "Open/reveal" and trash use `std::process::Command`
with argument vectors or native APIs; paths are passed as single arguments.
* Permanent delete (headless only) checks the item's owner and type against
the scan and removes it through an `os.Root` on the parent directory, so
a symlink swapped in since the scan cannot redirect it.
the scan and removes it through descriptors opened from the parent
directory without following symlinks (`openat`/`unlinkat` with
`O_NOFOLLOW`), so a symlink swapped in since the scan cannot redirect it.
* `--read-only` disables every mutating action in the UI.
DEVELOPMENT.md
+58 -44
View File
@@ -1,61 +1,77 @@
# Development
Requires Go (see `go.mod`). No CGO, no external tools.
Requires a stable Rust toolchain (`rustup`); the dependencies are pinned in
`Cargo.lock`. No C toolchain beyond what Rust itself needs.
```sh
make build # bin/mapsize, bin/treegen
make test # go test ./...
make race # go test -race ./...
make vet fmt-check
make benchmark # go test -bench
make fuzz # short fuzz runs of snapshot / treemap / filter
make release # release archives and packages in bin/release
make build # target/release/mapsize
make test # cargo test --release
make lint # clippy, warnings are errors
make fmt-check # rustfmt
make benchmark # in-process benchmarks (layout, render, filter): examples/bench.rs
make treegen # synthetic test-tree generator: examples/treegen.rs
make release # local release assets in dist/release (see below)
```
CI (`.github/workflows/ci.yml`) runs rustfmt and clippy, the tests on Linux
(x86-64 and ARM64), macOS (Intel and Apple silicon) and Windows, and
cross-builds every release target, running the ARM and RISC-V Linux
binaries under qemu.
## Platforms
`scripts/build.sh` builds `dist/mapsize-OS-ARCH`:
| OS | Architectures | Rust target | Built with |
|---|---|---|---|
| Linux | amd64, arm64, 386, arm (v7), riscv64 | `*-unknown-linux-musl` (static) | cargo-zigbuild + zig |
| Windows | amd64, arm64 | `x86_64-pc-windows-gnu`, `aarch64-pc-windows-gnullvm` | cargo-zigbuild + zig |
| macOS | amd64, arm64 | `*-apple-darwin` | cargo on a Mac (or zigbuild) |
| FreeBSD, NetBSD | amd64 | `x86_64-unknown-{freebsd,netbsd}` | cross + Docker |
Platform differences live in `src/platform.rs` (and `cfg` blocks where the
terminal code needs them); nothing else switches on the OS.
## Releasing
Example for 1.2.0; substitute the real version throughout.
1. Update the version in `internal/brand/brand.go` and `docs/mapsize.1`,
and add a `## 1.2.0` section to CHANGELOG.md. Commit.
1. Update the version in `Cargo.toml` (then `cargo check` to refresh
`Cargo.lock`) and `docs/mapsize.1`, and turn `## Unreleased` in
CHANGELOG.md into `## 1.2.0 — DATE`. Commit.
2. Tag and push:
```sh
git tag -a v1.2.0 -m "mapsize 1.2.0"
git push origin master v1.2.0
```
3. Build everything (archives, .deb/.rpm/.apk/Arch packages, SHA256SUMS
into `bin/release`; the first run downloads `nfpm`):
```sh
make release VERSION=v1.2.0
```
4. Publish the GitHub release with the changelog section as notes:
```sh
awk '/^## 1.2.0/{f=1;next} /^## /{f=0} f' CHANGELOG.md > /tmp/notes.md
gh release create v1.2.0 --title v1.2.0 --notes-file /tmp/notes.md bin/release/*
```
3. The tag starts `.github/workflows/release.yml`, which builds every
target, packages them (archives, .deb/.rpm/.apk/Arch packages,
SHA256SUMS) and publishes the GitHub release with the changelog section
as its notes. Watch it with `gh run watch`. `make release` builds the
same assets locally into `dist/release`, for whichever targets the host
can build (`scripts/build.sh`, then `scripts/release.sh`).
4. Check the release page once the workflow has finished.
5. Update Homebrew. This needs the tag pushed (step 2), because it
checksums GitHub's source tarball for that tag:
```sh
make brew-formula VERSION=v1.2.0 # writes bin/homebrew/mapsize.rb
make brew-formula VERSION=v1.2.0 # writes dist/homebrew/mapsize.rb
git clone https://github.com/andydixon/homebrew-tap /tmp/tap
cp bin/homebrew/mapsize.rb /tmp/tap/Formula/mapsize.rb
cp dist/homebrew/mapsize.rb /tmp/tap/Formula/mapsize.rb
git -C /tmp/tap commit -am "mapsize 1.2.0" && git -C /tmp/tap push
```
Users get it with `brew upgrade mapsize`. Template:
`packaging/homebrew/mapsize.rb.in`.
6. Publish to the apt, dnf and pacman repository at repo.dixon.cx. This
builds signed .deb, .rpm and Arch packages from the tag and rebuilds the
repository indexes (needs docker, gpg, apt-ftparchive and the signing key
in `~/.config/repo.dixon.cx/gnupg`):
repository indexes (needs cargo-zigbuild and zig, docker, gpg,
apt-ftparchive and the signing key in `~/.config/repo.dixon.cx/gnupg`):
```sh
packaging/repo.sh v1.2.0
```
The same script, with its own project block, publishes vault.
Always pass `VERSION=`: otherwise it comes from `git describe`, which only
matches the tag when HEAD is exactly the tagged commit. `make release` and
the Homebrew formula stamp the version into the binary; `make build` and
`go install` use the value in `brand.go`, which is why step 1 updates it.
The binary reports the version in `Cargo.toml`; `scripts/release.sh`
refuses to package a tag that does not match it.
The snapshot format is stable at version 1: a format change must bump
`snapshot.Version` and keep the reader for version 1.
@@ -65,17 +81,16 @@ The snapshot format is stable at version 1: a format change must bump
`treegen` creates synthetic trees:
```sh
bin/treegen --out /tmp/t --directories 50000 --files 1000000 --max-depth 20 --sparse --unicode
bin/mapsize /tmp/t
cargo run --release --example treegen -- --out /tmp/t --directories 50000 --files 1000000 --max-depth 20 --sparse --unicode
target/release/mapsize /tmp/t
```
## Benchmarks
```sh
make benchmark # Go micro-benchmarks (layout, render, filter, scan)
make benchmark # in-process: layout, render, navigation, filters
scripts/bench-scan.sh /path 1 2 4 8 # wall time and peak RSS per worker count
sudo DROP_CACHES=1 scripts/bench-scan.sh /path # cold-cache runs
MAPSIZE_MEM_ROOT=/path go test ./internal/scan -run TestMemoryPerNode -v
```
Paste results into PERFORMANCE.md with the machine description. Never edit
@@ -84,26 +99,25 @@ numbers by hand.
## Terminal restoration checks
```sh
MAPSIZE_DEBUG_PANIC=ui bin/mapsize . # panics on the first key press
MAPSIZE_DEBUG_PANIC=scanner bin/mapsize / # panics in a scanner goroutine
MAPSIZE_DEBUG_PANIC=ui target/release/mapsize . # panics on the first key press
MAPSIZE_DEBUG_PANIC=scanner target/release/mapsize / # panics in a scanner thread (aborts)
stty -a # echo/icanon must be back on
```
## Profiling
```sh
bin/mapsize --no-ui --cpuprofile cpu.prof --memprofile mem.prof /path
go tool pprof -http=: cpu.prof
```
`--log FILE --log-level debug` writes a log without disturbing the TUI.
Paths are logged only at `debug`/`trace` level.
Build with symbols (`CARGO_PROFILE_RELEASE_STRIP=false
CARGO_PROFILE_RELEASE_DEBUG=1 cargo build --release`) and use the platform
profiler, for example `perf record -g target/release/mapsize --no-ui /path`
on Linux or Instruments on macOS. `examples/bench.rs` times the hot UI paths
in-process.
## Layout
See ARCHITECTURE.md. Rules of thumb:
* `internal/treemap` and `internal/filter` must not import UI packages.
* Only `internal/platform` may contain OS-specific code (build tags).
* All strings derived from filenames go through `textutil.Sanitize` before
* `treemap` and `filter` must not depend on `tui`.
* OS-specific code lives in `platform` (`cfg` attributes); the terminal code
has the only other `cfg` blocks.
* All strings derived from filenames go through `textutil::sanitize` before
reaching a terminal.
PERFORMANCE.md
+69 -61
View File
@@ -3,107 +3,115 @@
All numbers below were measured, not estimated. Re-run them on your own
hardware before relying on them; filesystem caches dominate scan timings.
## Test machine (2026-10-02)
## Test machine (2026-10-08)
| | |
|---|---|
| CPU | Intel Core i7-7700 @ 3.60 GHz, 4 cores / 8 threads |
| RAM | 32 GiB |
| OS | Linux 7.0, Go 1.27.1 |
| OS | Linux 7.0, Rust 1.91.1 (release profile: LTO, one codegen unit) |
| Filesystem | ext4 on Linux software RAID (`/dev/md2`) over two 7200 rpm SATA HDDs |
The host is shared; its load average was 5–9 during these runs. Each figure
is the median (and best) of 7 runs.
## Scanner
Dataset: `~/go/pkg/mod` — 388,307 files + 65,518 directories (458,510
inventory nodes), 10.5 GiB allocated. **Warm cache only**: dropping the page
cache needs root, which was not available, so these measure CPU and syscall
Dataset: `~/go/pkg/mod`, a large module cache: 388,307 files + 65,518
directories (458,510 inventory nodes), 10.5 GiB allocated. **Warm cache
only**: dropping the page cache needs root, so these measure CPU and syscall
cost, not disk seeks. Cold-cache behaviour on HDDs will be dominated by seek
time; expect fewer workers to be optimal there.
`mapsize --no-ui --workers N ~/go/pkg/mod`, wall time, 3 runs each
(`/usr/bin/time`):
`mapsize --no-ui --workers N PATH`:
| workers | run 1 | run 2 | run 3 | entries/s (best) | peak RSS |
|---:|---:|---:|---:|---:|---:|
| 1 | 1.77 s | 1.67 s | 1.65 s | ~278,000 | 117 MB |
| 2 | 1.06 s | 0.99 s | 1.00 s | ~463,000 | 120 MB |
| 4 | 0.59 s | 0.59 s | 0.62 s | ~777,000 | 117 MB |
| 8 (balanced default) | 0.48 s | 0.50 s | 0.51 s | ~955,000 | 121 MB |
| 16 | 0.45 s | 0.44 s | 0.43 s | ~1,066,000 | 121 MB |
| 32 | 0.43 s | 0.44 s | 0.42 s | ~1,092,000 | 122 MB |
| workers | median | best | peak RSS |
|---:|---:|---:|---:|
| 1 | 1.52 s | 1.44 s | 75 MB |
| 8 (balanced default) | 0.47 s | 0.43 s | 76 MB |
| 32 | 0.46 s | 0.43 s | 80 MB |
For comparison, `du -s` on the same tree (warm): 1.59 s, 34 MB RSS.
At 8 workers that is about 1,000,000 entries a second, in under 80 MB for
458k nodes. Most of the time is spent in the `statx` and `getdents` system
calls.
CPU profile at 8 workers: ~60 % of samples in `statx`/`getdents` syscalls;
path reconstruction for job dispatch ~7 %; aggregation is negligible.
### Memory
`go test ./internal/scan -run TestMemoryPerNode` with
`MAPSIZE_MEM_ROOT=~/go/pkg/mod`: 458,510 nodes, 69.1 MiB live heap,
**~158 bytes per node** (node record ~112 bytes plus name storage and
directory side tables). Extrapolated, 10 M entries need ~1.6 GiB.
| other runs over the same tree | median | peak RSS |
|---|---:|---:|
| scan + `--largest-files 50` | 0.60 s | 76 MB |
| scan + `--json` (streamed to /dev/null) | 0.86 s | 76 MB |
## Snapshots
Same 458,510-node tree:
| operation | median | peak RSS |
|---|---:|---:|
| scan + save | 0.75 s | 77 MB |
| load (`mapsize --no-ui file.msz`) | 0.23 s | 78 MB |
| load a snapshot written by mapsize 1.2.0 | 0.26 s | 78 MB |
| operation | result |
|---|---:|
| snapshot file size (gzip level 1) | 6.7 MB (~15 bytes/node) |
| scan + save | 1.03 s |
| load (`mapsize --no-ui file.msz`), 3 runs | 0.28 / 0.29 / 0.32 s, 81 MB RSS |
| warm-cache rescan for comparison | 0.55 s |
Loading decompresses and verifies SHA-256 on a separate goroutine, overlapping
with parsing. Against a warm cache the win is ~1.9×; against a cold cache or
a network filesystem, where scans are dominated by I/O latency, it is far
larger (not yet measured here).
Loading decompresses and verifies SHA-256 on one thread while another parses.
Against a warm cache that is about twice as fast as rescanning; against a
cold cache or a network filesystem, where scans are dominated by I/O
latency, the gain is far larger (not yet measured here).
## UI
`go test ./internal/tui -bench Big` on a synthetic in-memory tree of
1,008,201 nodes (200 directories × 40 subdirectories × 100 files, plus one
flat directory with 200,000 files), 200×60 terminal:
`make benchmark` (`examples/bench.rs`), best of 5 runs, on a synthetic
in-memory tree of 1,008,201 nodes (200 directories × 40 subdirectories × 100
files, plus one flat directory with 200,000 files), 200×60 terminal:
| operation | time per op |
|---|---:|
| cold frame at root (sort children, LOD, layout incl. nested previews, paint) | 33.8 ms |
| resize to a new size (layout + paint, child lists cached) | 2.0 ms |
| arrow-key move + re-render | 1.5 ms |
| zoomed into the 200,000-entry directory, cold | 29.5 ms |
| render of a small tree (2.4 K nodes) at 200×60 | 0.66 ms |
| resize relayout of the small tree | 0.45 ms |
| cold frame at root (sort children, LOD, layout incl. nested previews, paint) | 11.34 ms |
| resize to a new size (layout + paint, child lists cached) | 1.50 ms |
| arrow-key move + re-render | 1.31 ms |
| zoomed into the 200,000-entry directory, cold | 9.37 ms |
During a scan the size caches are invalidated at most every 400 ms, so a
cold frame (worst case above ~34 ms) costs under 10 % of one core; other
frames reuse caches. Resize events only record the new size; layout happens
once per rendered frame, so a resize storm cannot queue up work.
During a scan the size caches are invalidated at most every 400 ms, so even
a cold frame costs a few percent of one core. Resize events only record the
new size; layout happens once per rendered frame, so a resize storm cannot
queue up work.
### Geometry
`go test ./internal/treemap -bench .`:
| operation | time |
|---|---:|
| squarified layout, 300 items | 89 µs |
| spatial neighbour lookup among 300 blocks | 2.1 µs |
| squarified layout, 300 items | 8.4 µs |
| spatial neighbour lookup among 300 blocks | 1.1 µs |
### Filtering (1,008,201 nodes)
| query | time |
|---|---:|
| `size > 1MB` | 31 ms |
| `*.dat AND size > 500k` | 119 ms |
| `path contains sub3` | 112 ms |
| `size > 1MB` | 26.75 ms |
| `*.dat AND size > 500k` | 67.84 ms |
| `path contains sub3` | 63.80 ms |
Filters run on a background goroutine with cancellation and a 120 ms
keystroke debounce; the UI never waits for them.
Filters run on a background thread with cancellation and a 120 ms keystroke
debounce; the UI never waits for them. Path predicates build each path from
the cached parent directory path into a reused buffer, and `*.ext` globs are
a suffix test.
## Compared with mapsize 1.2.0
mapsize 1.2.0 was the last release of the previous implementation, written
in Go. Both were run alternately on the same machine, over the same tree, in
the same session:
| | 1.2.0 | current | |
|---|---:|---:|---|
| scan, 8 workers (median) | 0.63 s | 0.47 s | 1.34× faster |
| scan, 1 worker (median) | 1.89 s | 1.52 s | 1.24× faster |
| peak RSS while scanning | 117 MB | 76 MB | |
| scan + save snapshot | 1.22 s | 0.75 s | 1.63× faster |
| load snapshot | 0.34 s | 0.23 s | 1.48× faster |
| scan + JSON export | 1.27 s, 153 MB | 0.86 s, 76 MB | 1.48× faster |
| cold frame, 1M nodes | 36.01 ms | 11.34 ms | 3.18× faster |
| squarified layout, 300 items | 84.3 µs | 8.4 µs | 10.07× faster |
| arrow key + re-render | 1.50 ms | 1.31 ms | 1.14× faster |
| filter `*.dat AND size > 500k` | 121.40 ms | 67.84 ms | 1.79× faster |
## Known costs / future work
* Path predicates now build each path from the cached parent directory path
(244 ms → 112 ms); globs are dominated by `filepath.Match`.
* Node records could shrink (owner/group/mode/link count are only needed for
display and some filters) if memory on 10 M+ entry trees matters.
* Cold-cache HDD benchmarks still need to be run with root access.
RUST-PORT-NOTES.md
-93
View File
@@ -1,93 +0,0 @@
# mapsize — Rust port
A port of the Go implementation (repo root) to Rust, with the same CLI, TUI,
keys, themes and snapshot format. `.msz` snapshots are byte-compatible in
both directions. The code is Linux/Unix only; the Windows code paths were
not ported.
```
cd rust
cargo build --release # target/release/mapsize
cargo test --release # 69 tests (ported from the Go suites, fuzz tests excepted)
```
Module layout mirrors `internal/`: `inventory`, `scan`, `platform`, `treemap`,
`filter`, `snapshot` (+ `compare`), `duplicate`, `export`, `config`,
`textutil`, `tui`, `app`. The tree is shared as `Arc<RwLock<Tree>>` with the
same single-writer invariants as ARCHITECTURE.md; `cancel::Cancel` replaces
`context.Context`; crossterm replaces Bubble Tea as the event loop (the
canvas renderer is ported as-is, plus line-diffed frame output).
## Known differences from Go
* `--log`/`--log-level` are accepted but there is no debug log yet.
* A scanner thread panic restores the terminal and aborts the process instead
of returning an "internal error" message.
* The colour profile is detected from `NO_COLOR`/`TERM`/`COLORTERM` only (no
terminal query); `--color` still overrides.
* JSON export does not HTML-escape `<>&`; `started` is written in UTC.
* The filter lexer no longer hangs on queries containing bytes Go treats as
white space but never skips (e.g. `voilà`, whose UTF-8 contains 0xA0).
* Huge sizes saturate rather than wrap in filters and snapshot comparison.
## Benchmarks
Machine: Intel Core i7-7700 (4C/8T), 32 GiB, ext4 on md RAID over 7200 rpm
HDDs. Go 1.27.1, Rust 1.91.1 (`lto`, `codegen-units = 1`). The host is
shared, and its load average was 5–9 on 8 threads during the runs. Go's
absolute times are close to those in PERFORMANCE.md. Go and Rust runs were
interleaved; the tables report best-of (micro) or median and best
(end-to-end). To re-run:
```
go test -c -o /tmp/tui.test ./internal/tui && go test -c -o /tmp/treemap.test ./internal/treemap
(cd rust && cargo build --release --example bench)
rust/bench/micro.py /tmp/tui.test /tmp/treemap.test rust/target/release/examples/bench 5
rust/bench/compare.py bin/mapsize rust/target/release/mapsize ~/go/pkg/mod 7
```
### In-process (Go `go test -bench` suite vs `examples/bench.rs`)
The tree is 1,008,201 synthetic nodes, rendered at 200×60.
| benchmark | Go | Rust | speed-up |
|---|---:|---:|---:|
| treemap/Squarify(300) | 84.3 µs | 8.4 µs | 10.07× |
| treemap/Neighbour(300) | 2.1 µs | 1.1 µs | 1.94× |
| tui/BigColdFrame | 36.01 ms | 11.34 ms | 3.18× |
| tui/BigResize | 2.02 ms | 1.50 ms | 1.35× |
| tui/BigFlatDirCold | 32.00 ms | 9.37 ms | 3.41× |
| tui/BigNavigate | 1.50 ms | 1.31 ms | 1.14× |
| tui/BigFilter/size > 1MB | 29.44 ms | 26.75 ms | 1.10× |
| tui/BigFilter/*.dat AND size > 500k | 121.40 ms | 67.84 ms | 1.79× |
| tui/BigFilter/path contains sub3 | 101.86 ms | 63.80 ms | 1.60× |
### End to end (`~/go/pkg/mod`: 388k files + 65k dirs, warm cache)
| scenario | Go median | Rust median | Go best | Rust best | Go RSS | Rust RSS | speed-up |
|---|---:|---:|---:|---:|---:|---:|---:|
| scan, 1 workers | 1.89 s | 1.52 s | 1.72 s | 1.44 s | 112 MB | 75 MB | 1.24× |
| scan, 8 workers | 0.63 s | 0.47 s | 0.58 s | 0.43 s | 117 MB | 76 MB | 1.34× |
| scan, 32 workers | 0.60 s | 0.46 s | 0.48 s | 0.43 s | 118 MB | 80 MB | 1.30× |
| scan + save snapshot | 1.22 s | 0.75 s | 1.13 s | 0.71 s | 119 MB | 77 MB | 1.63× |
| load own snapshot | 0.34 s | 0.23 s | 0.30 s | 0.23 s | 85 MB | 78 MB | 1.48× |
| load Go snapshot | 0.36 s | 0.26 s | 0.30 s | 0.23 s | 85 MB | 78 MB | 1.38× |
| scan + JSON export | 1.27 s | 0.86 s | 1.04 s | 0.68 s | 153 MB | 76 MB | 1.48× |
| scan + top 50 largest files | 0.70 s | 0.60 s | 0.64 s | 0.47 s | 114 MB | 76 MB | 1.17× |
Reading the numbers:
* **Scanning** is 1.2–1.3× faster at every worker count. It is syscall-bound
(`statx`/`getdents`), so the gain comes from lower per-entry overhead in the
controller (no GC, no per-chunk allocation churn), not from faster I/O.
* **Memory**: peak RSS is about a third lower (76 MB vs 117 MB while
scanning 458k nodes). The JSON export halves memory (76 MB vs 153 MB) because
it streams without per-node allocations.
* **Snapshots**: scan + save is 1.6× faster and load is 1.4–1.5× faster. The
two binaries load each other's files.
* **UI**: cold frames are 3.2–3.4× faster, and squarified layout is about 10×
faster. Navigation and resize gain less (1.1–1.35×) because they mostly
reuse cached layout.
* **Filters**: 1.1–1.8× faster. This required caching lower-cased directory
paths in a `Vec` with reused buffers, and a `*.ext` fast path in the
`filepath.Match` port.
scripts/bench-scan.sh
+1 -1
View File
@@ -10,7 +10,7 @@ set -euo pipefail
path=${1:?usage: bench-scan.sh PATH [workers...]}
shift || true
workers=${*:-1 2 4 8 16}
bin=${MAPSIZE:-bin/mapsize}
bin=${MAPSIZE:-target/release/mapsize}
[[ -x $bin ]] || make build >/dev/null
fs=$(df -T "$path" | awk 'NR==2 {print $2}')