Compression foundations: what can shrink and what cannot

Reduction is not magic. It works by finding repetition and by accepting a small, controlled loss. Knowing which of the two a tool performs on which kind of data tells you how much room you can honestly expect to reclaim.

Two families of method

The first family is lossless. It records the same information in fewer bytes by describing patterns instead of listing every value. When the data is restored, it is byte-for-byte identical to the original. The saving depends entirely on how repetitive the data is. A text file full of repeated phrases shrinks dramatically. A file that has already been compressed shrinks hardly at all.

The second family is lossy. It discards detail that a human observer is unlikely to notice. Audio and image formats often use it, and the saving can be large. The trade is that the original cannot be recovered exactly, and repeated passes degrade quality quickly. For a game library, lossy methods belong to the production stage, not to day-to-day housekeeping.

What compresses well

Three kinds of data reward lossless reduction. Structured text and tables, such as saved data and configuration, compress extremely well. Uncompressed media, such as a raw audio bed or a raw bitmap, compresses moderately to well. And large collections of similar files compress well as a group, because the tool can share one dictionary across the whole set instead of starting fresh with each file.

What barely moves

Most of a modern release is already compressed. Textures arrive in a packed format, audio arrives in a packed format, and video arrives in a packed format. Running a general-purpose compressor over an already-packed directory yields only a few percent, occasionally less. That is not a failure of the tool; it is a statement about the data.

Where the honest wins are

The real gains come from the sources described in our anatomy of a footprint. Duplicated assets can sometimes be consolidated. Optional high-resolution packs can be omitted on a display that cannot show them. Language sets and bonus material can be trimmed to what is actually used. Each of those is a decision about content, not a clever trick, and together they can return more room than any general-purpose compressor used on the same directory.

Choosing the right lever

A useful rule of thumb: if a directory is rarely touched and holds loose or structured data, a general-purpose compressor is worth running. If it holds a packing-heavy library, the compressor is mostly a formality. In both cases the operating system already provides a suitable tool, and the following pages describe the facilities for Windows and for macOS and Linux. Nothing beyond the built-in options is required, and we recommend nothing beyond them.