DevZIP

DevZIP is a Windows-first archival project built around a native C++ engine and a readable .dvz container format.

Yes, .dvz is one letter away from DBZ, so a few power-level jokes were inevitable.

The project is focused on size-first compression for mixed real-world datasets. The current engine combines deterministic archive metadata, transform preprocessing, solid packing, deduplication, and pluggable native compression backends so the archive format can evolve without becoming unreadable.

Status

• Shipping native lane: best-of-two
• Archive format: readable .dvz
• Native CLI: compress, extract, verify
• Desktop app: WPF shell in apps/windows-ui/DevZip.App
• Public repo note: large generated corpora and benchmark output are intentionally excluded to keep the repository lightweight

Why DevZIP Exists

DevZIP is trying to answer a practical question: can a custom archival container beat familiar desktop archivers on mixed corpora while still staying commercially usable and internally maintainable?

The current answer is promising:

• The shipping native lane beats 7z-lzma2 on the latest mixed-large aggregate rerun.
• The archive remains self-describing by stamping the backend identity into the manifest.
• Experimental backend lanes still produce readable .dvz archives and can be compared without changing the container format.
• The current shipping lane is the stable "final form." The experimental ones are still in the gravity chamber.

Top Victories

If the goal is to show where DevZIP looks strongest today, these are the headline wins of the current shipping lane against the project's main size baseline, 7z-lzma2.

Category	DevZIP (MB)	7z-lzma2 (MB)	Win
Text and structured data	0.513	0.762	-32.67%
Raw bitmaps	59.244	77.786	-23.84%
Mixed-large aggregate	143.815	144.841	-0.71%
JPEG photographs	79.842	81.095	-1.54%
PNG lossless images	15.870	16.021	-0.95%
Video (raw YUV)	1.854	1.871	-0.94%
Random / high-entropy	10.486	10.487	-0.01%

Those are the strongest current wins. The only category still not clearly ahead in the shipping lane is software binaries, where DevZIP is near-tied at +0.32%; the experimental selective-zpaq5 lane is currently the strongest measured code-heavy follow-up.

Full Aggregate Comparison

Latest shipping-lane aggregate result from mixed-large:

Tool	Status	End Size (MB)	Total Time (s)	Delta vs 7z-lzma2
devzip-native	ok	143.815	304.2	-0.71%
7z-lzma2	ok	144.841	107.2	+0.00%
7z-lzma	ok	144.846	107.1	+0.00%
winrar	ok	148.261	48.9	+2.36%
7z-ppmd	ok	149.306	86.1	+3.08%
7z-bzip2	ok	160.398	265.2	+10.74%
windows-zip	ok	181.973	13.8	+25.64%

Notes:

• Lower end size is better.
• The 7z-deflate lane is omitted here because the latest aggregate run was partial, not a clean apples-to-apples result.
• Weissman scoring is still tracked internally, but DevZIP is gated primarily on end size.
• The table above is the full aggregate view. The section above it is the curated "greatest hits" version.
• The goal is smaller archives, not yelling until the compression ratio goes over 9000.

Per-Type Highlights

Against 7z-lzma2, the current shipping native lane reports:

• Text and structured data: -32.67%
• Raw bitmaps: -23.84%
• JPEG: -1.54%
• PNG: -0.95%
• Video: -0.94%
• Random data: -0.01%
• Software binaries: near-tied at +0.32%

Latest Native Backend Experiments

All rows below are native-only backend variants. They all produce readable .dvz files.

Dataset	Backend	End Size (MB)	Total Time (s)	Delta vs best-of-two
large-text	best-of-two	175.432	516.8	baseline
large-text	best-of-three-ppmd	175.432	556.8	+3 bytes
large-text	selective-zpaq5	175.432	521.6	+4 bytes
large-code	best-of-two	151.548	486.3	baseline
large-code	best-of-three-ppmd	151.212	525.5	-0.22%
large-code	selective-zpaq5	150.442	535.8	-0.73%

Current takeaway:

• best-of-three-ppmd does not help the large text corpus in this pipeline, but it does help code-heavy data a little.
• selective-zpaq5 is the strongest measured code-heavy candidate so far. On large-code, it currently has the highest scouter reading.
• mixed-large still needs a full rerun before the shipping default should change, so this is not the Super Saiyan switchover yet.

What Ships Today

Native engine

Located in native/engine.

Key traits:

• Deterministic .dvz manifest and block layout
• Transform pipeline with preprocessing and chunking
• Solid packing by file type
• Full-file and block-level deduplication
• Backend stamping for forward-readable archives
• Native backend selection through the CLI

Current CLI backends:

• best-of-two (default)
• best-of-three-ppmd
• selective-zpaq5
• lzma2
• ppmd
• libzpaq-4
• libzpaq-5

Windows desktop app

Located in apps/windows-ui/DevZip.App.

This is a WPF shell that delegates compression and extraction work to devzip_cli.exe.

Docs and benchmark harness

• Format docs: docs/format/dvz-format.md
• Benchmark notes: docs/benchmarks/baseline-results.md
• Licensing boundary notes: docs/specs/licensing.md
• Benchmark harness: benchmarks/

Building

Native engine and CLI

cmake -S native/engine -B native/engine/build
cmake --build native/engine/build --config Release --target devzip_cli devzip_tests

Run the native tests:

.\native\engine\build\devzip_tests.exe

Windows UI

dotnet build apps/windows-ui/DevZip.App/DevZip.App.csproj

CLI Usage

Basic compression:

.\native\engine\build\devzip_cli.exe compress "path\to\folder" "archive.dvz"

Select an experimental backend:

.\native\engine\build\devzip_cli.exe --backend selective-zpaq5 compress "path\to\folder" "archive.dvz"

Extract:

.\native\engine\build\devzip_cli.exe extract "archive.dvz" "output-folder"

Verify:

.\native\engine\build\devzip_cli.exe verify "archive.dvz"

Benchmarking

Run the benchmark harness:

python benchmarks/run_benchmarks.py --manifest benchmarks/manifests/mixed-large.json

Write the Markdown report too:

python benchmarks/run_benchmarks.py --manifest benchmarks/manifests/mixed-large.json --markdown-out docs/benchmarks/baseline-results.md

Important:

• sample-data/ is not included in this public repository because the generated corpora are very large.
• BenchmarkResults/ is also excluded because it contains generated output, ad hoc experiments, and large archive artifacts.
• The benchmark scripts and manifests remain in the repo so the datasets and reports can be regenerated locally.

Repository Layout

Path	Purpose
apps/windows-ui/DevZip.App	WPF shell
native/engine	Native C++ engine, CLI, tests, vendored codec deps
benchmarks	Corpus manifests, wrappers, and benchmark harness
docs/format	.dvz format notes
docs/specs	Design and licensing notes
python_tests	Python-side test helpers
README.html / README.css	Visual benchmark dashboard

Licensing

This repository is source-available under the custom DevZIP Non-Commercial License 1.0.

• Non-commercial use, study, modification, and redistribution are allowed under the license terms.
• Commercial use requires a separate written commercial license.
• Commercial licensing contact: contact@mikethetech.com

Important:

• This is not an OSI-approved open source license.
• It is also not a GNU license, because GNU licenses permit commercial use.
• Vendored third-party components keep their own upstream license terms. See docs/specs/licensing.md for the project boundary notes.

Commercial Licensing

For commercial licensing, business use, OEM bundling, internal enterprise deployment, paid services, or other revenue-linked usage, contact:

contact@mikethetech.com