RAR, Eugene Roshal, and the Licence That Splits Decoder from Encoder

The paragraph that decides everything

The load-bearing document in this story is not a specification, a court filing or an interview. It is a licence file roughly a page long, distributed inside RARLAB's own UnRAR source tarball, which anybody can download without agreeing to anything first. Its first clause reads: "All copyrights to RAR and the utility UnRAR are exclusively owned by the author - Alexander Roshal." Its second clause is the one that shaped how RAR-handling software could be built.

Read that as an engineering instruction rather than as legal boilerplate. Everything a program needs in order to decompress a RAR archive is handed over: real source code, usable in any software, free of charge, without limitation. Everything a program would need in order to compress one is withheld, and withheld specifically — not by silence or obscurity, but by an explicit prohibition on building a compatible archiver or re-creating the algorithm. A reader can be built by anyone. A writer cannot be built legitimately at all.

The consequence is the thing most people notice about RAR without ever knowing why. The Library of Congress, assessing the format for preservation purposes, states it flatly: RAR files "are the native format for WinRAR software and can only be created through this tool which is licensed to win.rar GmbH although there are several options to open RAR files." That is not a description of a difficult format. It is a description of a licence.

How the split is maintained in practice

It would be reasonable to assume the decoder is a separate, hand-maintained project — a stripped reimplementation kept in step with the real one. It is not. The readme shipped alongside the licence explains that "Unrar source is subset of RAR and generated from RAR source automatically, by a small program removing blocks like #ifndef UNRAR ... #endif." The published decoder is the private codebase with the encoder mechanically cut out of it, marked off in advance by preprocessor guards so the cut can be repeated on demand.

This detail matters more than it first appears. It means the asymmetry is not an accident of history that nobody got round to fixing, and not a matter of the compressor being too entangled to release. The two halves live in one tree, and the boundary between what is given away and what is kept is drawn deliberately, in source, and re-drawn every time a new version ships. The line between decoder and encoder is a maintained artefact.

Who wrote RAR, and who owns it

Almost every account of this format says Eugene Roshal owns RAR. The primary documents say otherwise, and they say it twice over. The UnRAR licence vests all copyright in Alexander Roshal and is signed "Alexander L. Roshal." Independently, the Library of Congress records the copyright in "RAR's compression applications and libraries to Alexander Roshal, brother of Eugene Roshal," while naming Eugene as the person the format is named after. Two unrelated sources, one conclusion: the author of the code and the holder of the rights are different people.

There is an oddity here worth flagging rather than tidying away. The licence describes Alexander as "the author" — a word that sits awkwardly against every other account, including the Library of Congress entry, in which Eugene is the one who wrote the thing. The sources consulted for this page do not reconcile that, and this archive is not going to reconcile it by choosing whichever reading reads more smoothly. What can be stated is the division of roles the documents actually support.

It also follows that calling RARLAB "Roshal's company" without qualification overstates the record. win.rar GmbH is a German company holding a licence, with its own management, and the documents describe it as licensee rather than as author or owner.

The dates that will not confirm

Biographical writing about RAR is unusually confident for how little it rests on. The commonly repeated release dates — RAR in 1993, WinRAR in the spring of 1995 — trace to encyclopedias and vendor-adjacent pages rather than to any archival artefact, dated press release or contemporaneous trade item located for this page. The same is true of Roshal's reported birth date and birthplace. None of that means the dates are wrong. It means they are unverified, and an archive that repeats them as though they were settled is quietly manufacturing certainty.

One date does survive the test, and it belongs to a different program. Far Manager — the other program Roshal is widely known for — carries a BSD 3-Clause licence file reading "Copyright (c) 1996, Eugene Roshal" and "Copyright (c) 2000, Far Group." That is a primary document: a licence header, in a live repository, naming both the author and the year. It is the firmest dated fact in this biography, and it comes from the project he gave away rather than the one he did not.

A documented container with an undocumented compressor

"RAR is undocumented" is another convenient half-truth. For the current generation it is simply false: RARLAB publishes a technical note describing RAR 5.0's data structures in implementable detail — the archive signature, the general block format, variable-length integers, the main archive header, file and service headers, encryption headers, the end-of-archive marker. Anyone can read it. What that document does at the top, though, is tell you where it stops: if you need information about the algorithms, it says, use the UnRAR source code.

That single redirection is the licence expressed as documentation policy. The container — how blocks are framed, where headers sit, how the file is laid out — is published prose. The compression itself is published only as a decoder you may read and use but may not use to build a compressor. A preservation engineer can parse a RAR archive's structure from the vendor's own text; the moment they want to produce one, the text hands them back to the licence.

The earlier generations are a different problem, and the honest position on them is silence. The Library of Congress states that RAR 1.3 and RAR 1.5 have no public documentation at all and are only thought to underlie later versions. When the national library's format registry declines to describe a format's internals, no secondary account is in a position to do better; claims about what the first RAR versions did internally should be read as reconstruction, not record.

What is actually being withheld

It is tempting to imagine the withheld compressor as a piece of unshared mathematics. RARLAB's own acknowledgements file argues against that reading. It credits the text-compression side of the RAR4 generation to Dmitry Shkarin's PPMII work and to Dmitry Subbotin's carryless range coder, and credits the recovery-record functionality to published Reed-Solomon research. Those are named, public contributions from the compression literature, acknowledged by the vendor in the same source distribution that carries the licence.

So the thing behind clause 2 is better understood as an engineered combination — parameter choices, modelling, format integration, years of tuning — than as a secret unknown to the field. That is exactly why the restriction has to be written down rather than merely relied upon. Obscurity would not have held. A licence does.

The counterfactual next door

ZIP, a format from the same era, faced the same problem and made the opposite choice. PKWARE published a container specification openly and kept republishing it, and the Deflate method at ZIP's core was later specified independently at the IETF as RFC 1951 in May 1996, which left anyone who wanted one with an implementable description of the container from PKWARE and of the Deflate method from the IETF. The Library of Congress describes ZIP as a de facto industry standard, openly documented by its vendor. The outcome is the one everybody lives inside: a ZIP archive can be both read and written by software built from a freely published specification, without negotiating with the vendor — the classic Deflate core, at least; PKWARE reserves certain later patented technology behind a separate licence.

Set the two side by side and the variable isolates itself. Same era, same job, comparable technical seriousness, and one difference that matters — whether the compressor was made implementable by third parties. This page takes no position on which format compresses better, because that question is benchmark-dependent and, for this argument, beside the point. Universality did not follow the better compressor. It followed the published one.

The same author's other program went the other way

The contrast is not a claim about a person's disposition, and it would be lazy to make it one. Far Manager, the other program Roshal is known for, is published under a BSD 3-Clause licence — the permissive, OSI-recognised kind, with no clause carving out any part of the program from reuse — and the project's own site documents its move to published source after it passed out of his hands. One program by the same author is free software by any standard definition. The other publishes a decoder and prohibits an encoder.

What separates them is not temperament but commercial structure. Far Manager had no product on the other side of it; RAR did, and still does, in the form of a licensed archiver sold by a company that holds the rights under licence from the copyright holder. Clause 2 is the load-bearing wall of that arrangement. Remove it and the commercial position of the archiver changes, which is a plausible reading of why the clause is still there in the current source drop.

The clause, running on a phone

The reason this piece of licence archaeology is worth a reader's time is that its consequence is testable in about ten seconds on whatever device is nearest. Long-press a RAR archive on a phone and something will offer to extract it — because RARLAB's decoder may be used in any software free of charge. Try to produce one and the option is generally missing. ZIP & RAR, an archiver built for phones, sits exactly on that boundary: it can extract a RAR archive because the decoder is published for that purpose, and it can create ZIP archives because ZIP's compression is independently specified and implementable by anyone. The capability gap between those two operations is not a gap in the software. It is clause 2, executing.

That is also the useful lesson for anyone choosing a format for material meant to outlive the tool that made it. Both public format registries that catalogue RAR flag its proprietary status, and preservation guidance generally prefers containers with independent implementations for exactly the reason this page has been tracing: a format you can only write with one vendor's program is a format whose future is that vendor's to decide. RAR is not a cautionary tale about bad engineering. It is a demonstration that the terms attached to a format do more to determine its reach than the format's own quality ever will.