Merge pull request #16 from kduma-OSS/claude/dynamic-nonlinear-partitions-vSXIn

Author: kduma

.github/workflows/ci.yml

@@ -135,3 +135,30 @@ jobs:
       - run: cargo clippy -p pcf-compact --all-targets -- -D warnings
       - run: cargo build -p pcf-compact --verbose
       - run: cargo test -p pcf-compact --verbose
+
+  pcf-dcp:
+    name: pcf-dcp profile
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: .
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          components: rustfmt, clippy
+      - uses: Swatinem/rust-cache@v2
+      - run: cargo fmt -p pcf-dcp -- --check
+      - run: cargo clippy -p pcf-dcp --all-targets -- -D warnings
+      - run: cargo build -p pcf-dcp --verbose
+      - run: cargo test -p pcf-dcp --verbose
+      - name: Regenerate the spec test vector
+        run: cargo run -p pcf-dcp --example gen_testvector -- pcf_dcp_testvector.bin
+      - name: Inspect generated test vector (spec Section 17 is 700 bytes)
+        run: |
+          ls -l pcf_dcp_testvector.bin
+          test "$(wc -c < pcf_dcp_testvector.bin)" = "700"
+      - uses: actions/upload-artifact@v4
+        with:
+          name: pcf-dcp-testvector
+          path: pcf_dcp_testvector.bin

.github/workflows/release-prepare.yml

@@ -76,13 +76,16 @@ jobs:
           sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' reference/PCF-v1.0/Cargo.toml
           sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' reference/PFS-MS-v1.0/Cargo.toml
           sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' reference/PCF-SIG-v1.0/Cargo.toml
+          sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' reference/PCF-DCP-v1.0/Cargo.toml
           sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' tools/pcf-debug/Cargo.toml
           sed -i 's/^version = "[^"]*"/version = "'"$NEW"'"/' tools/pcf-compact/Cargo.toml
           # path-dep version pins on pcf
           sed -i 's|pcf = { path = "\.\./PCF-v1.0", version = "[^"]*" }|pcf = { path = "../PCF-v1.0", version = "'"$NEW"'" }|' reference/PFS-MS-v1.0/Cargo.toml
           sed -i 's|pcf = { path = "\.\./PCF-v1.0", version = "[^"]*" }|pcf = { path = "../PCF-v1.0", version = "'"$NEW"'" }|' reference/PCF-SIG-v1.0/Cargo.toml
+          sed -i 's|pcf = { path = "\.\./PCF-v1.0", version = "[^"]*" }|pcf = { path = "../PCF-v1.0", version = "'"$NEW"'" }|' reference/PCF-DCP-v1.0/Cargo.toml
           sed -i 's|pcf = { path = "\.\./\.\./reference/PCF-v1.0", version = "[^"]*" }|pcf = { path = "../../reference/PCF-v1.0", version = "'"$NEW"'" }|' tools/pcf-debug/Cargo.toml
           sed -i 's|pcf-sig = { path = "\.\./\.\./reference/PCF-SIG-v1.0", version = "[^"]*" }|pcf-sig = { path = "../../reference/PCF-SIG-v1.0", version = "'"$NEW"'" }|' tools/pcf-debug/Cargo.toml
+          sed -i 's|pcf-dcp = { path = "\.\./\.\./reference/PCF-DCP-v1.0", version = "[^"]*" }|pcf-dcp = { path = "../../reference/PCF-DCP-v1.0", version = "'"$NEW"'" }|' tools/pcf-debug/Cargo.toml
           sed -i 's|pcf = { path = "\.\./\.\./reference/PCF-v1.0", version = "[^"]*" }|pcf = { path = "../../reference/PCF-v1.0", version = "'"$NEW"'" }|' tools/pcf-compact/Cargo.toml
 
       - name: Bump TypeScript packages

.github/workflows/release.yml

@@ -132,6 +132,19 @@ jobs:
         if: needs.resolve.outputs.dry_run != 'true'
         run: sleep 45
 
+      - name: cargo publish pcf-dcp
+        shell: bash
+        run: |
+          if [ "${{ needs.resolve.outputs.dry_run }}" = "true" ]; then
+            cargo publish -p pcf-dcp --allow-dirty --dry-run
+          else
+            cargo publish -p pcf-dcp --allow-dirty --token "${{ steps.cargo-auth.outputs.token }}"
+          fi
+
+      - name: Wait for crates.io index
+        if: needs.resolve.outputs.dry_run != 'true'
+        run: sleep 45
+
       - name: cargo publish pcf-debug
         shell: bash
         run: |

Cargo.toml

@@ -4,6 +4,7 @@ members = [
     "reference/PCF-v1.0",
     "reference/PFS-MS-v1.0",
     "reference/PCF-SIG-v1.0",
+    "reference/PCF-DCP-v1.0",
     "tools/pcf-debug",
     "tools/pcf-compact",
     "tools/pcf-sig",

reference/PCF-DCP-v1.0/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "pcf-dcp"
+version = "0.0.8"
+edition = "2021"
+description = "Reference implementation of PCF-DCP v1.0, the PCF Dynamic Container Partition profile"
+license = "MIT OR Apache-2.0"
+repository = "https://github.com/kduma-OSS/Partitioned-Container-Format"
+homepage = "https://github.com/kduma-OSS/Partitioned-Container-Format"
+readme = "README.md"
+keywords = ["pcf", "dcp", "container", "deduplication", "fragmentation"]
+categories = ["encoding", "filesystem"]
+
+# This crate is a *reference* implementation of the PCF-DCP profile. Like the
+# `pcf` crate it builds on, it favours a direct, auditable mapping onto the
+# written specification (`specs/PCF-DCP-spec-v1.0.txt`) over raw performance.
+
+[[bin]]
+name = "dcp"
+path = "src/bin/dcp.rs"
+
+[dependencies]
+# The PCF-DCP profile is layered strictly above PCF v1.0; every byte container
+# operation goes through the reference PCF crate. The arena reuses PCF's Table
+# Block, Partition Entry, and table-hash primitives directly.
+pcf = { path = "../PCF-v1.0", version = "0.0.8" }

reference/PCF-DCP-v1.0/README.md

@@ -0,0 +1,120 @@
+# pcf-dcp — PCF Dynamic Container Partition (reference implementation)
+
+Reference reader/writer for **PCF-DCP v1.0**, an application-level profile that
+adds *dynamic*, fragmentable, dedup-friendly sub-partitions to the
+[Partitioned Container Format](../PCF-v1.0) without modifying the PCF byte
+container.
+
+This crate mirrors the written specification (`specs/PCF-DCP-spec-v1.0.txt`)
+field-for-field and is intended as the *normative* implementation against which
+language ports are checked. It favours auditability over performance.
+
+## Model at a glance
+
+PCF-DCP defines one new PCF partition type:
+
+| Type         | Name            | Holds                                              |
+|--------------|-----------------|----------------------------------------------------|
+| `0xAAAC0001` | `DCP_CONTAINER` | An *arena*: a header, an inner partition table, fragment tables, and data extents |
+
+A DCP container's bytes are an **arena** addressed by arena-relative offsets:
+
+```
+arena:
+[ DCP Header (24 B) | data extents | Fragment Tables | Inner Table Block(s) ]
+```
+
+* **DCP Header** — `"PDCP"` magic, profile version, `inner_table_offset`,
+  `arena_used` (a bump pointer).
+* **Inner Table Block** — a chain of reused PCF Table Blocks (74 B header +
+  141 B entries), byte-for-byte identical to the top-level table, listing the
+  *inner* partitions. Two entry fields are reinterpreted: `start_offset` points
+  at the partition's Fragment Table, and `max_length` equals `used_bytes`.
+* **Fragment Table** — per inner partition, a chain of 9-byte block headers each
+  followed by 18-byte **Fragment Entries**. Each entry names one extent
+  `(offset, length, kind, flags)`. The logical content of an inner partition is
+  the concatenation of its DATA extents.
+
+A generic PCF reader sees a DCP file as **one opaque partition**; only a
+DCP-aware reader looks inside. A DCP file is always a conforming PCF v1.0 file.
+
+## Why a profile
+
+PCF stores each partition as a contiguous, statically-reserved region. PCF-DCP
+makes each *inner* partition grow, shrink, and be edited in the middle without
+relocating its neighbours, by describing it as a list of extents rather than one
+range. This buys:
+
+* **Fragmentation / random edits** — append, insert, overwrite, delete, and
+  truncate are edits of the Fragment Table (copy-on-write for shared bytes); no
+  data is moved.
+* **Deduplication** — two extents may name the same arena bytes; identical
+  chunks are stored once. The per-extent `SHARED` flag makes safe in-place
+  editing explicit.
+* **Hash / signature stability** — an inner partition's `data_hash` covers its
+  *logical content*, so fragmentation, dedup, compaction, and promotion all
+  leave the hash (and any PCF-SIG signature over it) unchanged.
+
+## Library example
+
+```rust
+use std::io::Cursor;
+use pcf_dcp::{Arena, Chunker, DcpReader, DcpWriter, HashAlgo};
+
+let mut arena = Arena::new();
+arena.add_inner(0x10, [0xA1; 16], "A", b"Hello, World!", HashAlgo::Sha256, Chunker::Fixed(7))?;
+arena.add_inner(0x10, [0xB2; 16], "B", b"World!", HashAlgo::Sha256, Chunker::Whole)?;
+
+let mut w = DcpWriter::new();
+w.add_container([0xDC; 16], "dcp", arena)?;
+let image = w.to_image()?;
+
+let mut r = DcpReader::open(Cursor::new(image))?;
+r.verify()?;
+assert_eq!(r.read_inner(&[0xB2; 16])?, b"World!");
+# Ok::<(), pcf_dcp::Error>(())
+```
+
+## Promotion / demotion
+
+`DcpWriter::promote` moves an inner partition out to a top-level PCF partition
+(dynamic → fixed); `demote` moves a top-level partition into a container
+(fixed → dynamic). Both preserve `uid`, `partition_type`, `label`,
+`data_hash_algo_id`, and `data_hash` — the **promotion invariant**, identical to
+the set of fields a PCF-SIG signature protects.
+
+## Command-line tool
+
+The `dcp` binary inspects and rewrites DCP files; every mutating command
+re-verifies before writing:
+
+```
+dcp info    <file>
+dcp dedup   <file> [--fixed N] [--trailer]
+dcp defrag  <file> [--trailer]
+dcp promote <file> <container-uid> <inner-uid> [--trailer]
+dcp demote  <file> <part-uid> <container-uid> [--trailer]
+```
+
+UIDs are 32 hex digits, or `0xNN` for a uid of 16 identical bytes (e.g. `0xDC`).
+
+## Build & test
+
+```
+cargo test -p pcf-dcp
+cargo run -p pcf-dcp --example gen_testvector -- /tmp/dcp.bin   # the 700-byte vector
+cargo run -p pcf-dcp --bin dcp -- info /tmp/dcp.bin
+```
+
+The example reproduces the byte-exact 700-byte test vector from Section 17 of
+the specification.
+
+## Relationship to `pcf`
+
+This crate is layered strictly above [`pcf`](../PCF-v1.0): every container byte
+operation goes through the reference PCF crate, and the arena reuses PCF's Table
+Block, Partition Entry, and table-hash primitives directly.
+
+## Licence
+
+MIT OR Apache-2.0.

reference/PCF-DCP-v1.0/examples/gen_testvector.rs

@@ -0,0 +1,57 @@
+//! Generates the canonical PCF-DCP v1.0 test-vector file used in spec
+//! Section 17.
+//!
+//! Run with: `cargo run --example gen_testvector -- <output-path>`
+//! (defaults to ./pcf_dcp_testvector.bin). Everything is fixed and
+//! deterministic so that ports can reproduce the file byte-for-byte.
+
+use std::io::Cursor;
+
+use pcf::Container;
+use pcf_dcp::{build_reference_vector, DcpReader};
+
+fn main() {
+    let path = std::env::args()
+        .nth(1)
+        .unwrap_or_else(|| "pcf_dcp_testvector.bin".to_string());
+
+    let image = build_reference_vector().expect("build reference vector");
+    std::fs::write(&path, &image).expect("write file");
+
+    // It is a conforming PCF v1.0 file ...
+    let mut pcf = Container::open(Cursor::new(image.clone())).expect("pcf open");
+    pcf.verify().expect("pcf verify");
+
+    // ... and a conforming DCP file.
+    let mut dcp = DcpReader::open(Cursor::new(image.clone())).expect("dcp open");
+    dcp.verify().expect("dcp verify");
+
+    eprintln!("wrote {} ({} bytes)", path, image.len());
+    for c in dcp.containers().expect("containers") {
+        let arena = dcp.open_arena(&c).expect("arena");
+        eprintln!(
+            "  container {:<6} type=0x{:08X} used={} inners={}",
+            c.label_string().unwrap_or_default(),
+            c.partition_type,
+            c.used_bytes,
+            arena.len()
+        );
+        for info in arena.inners() {
+            let n = info.data_hash_algo.digest_len();
+            let hex: String = info.data_hash[..n]
+                .iter()
+                .map(|b| format!("{b:02x}"))
+                .collect();
+            let shared = info.extents.iter().filter(|e| e.shared).count();
+            eprintln!(
+                "    inner {:<3} type=0x{:08X} used={} extents={} shared={} data_hash={}",
+                info.label,
+                info.partition_type,
+                info.used_bytes,
+                info.extents.len(),
+                shared,
+                hex
+            );
+        }
+    }
+}