Encapsulation

Raw DEFLATE compressed data (RFC 1951)^[5] are typically written with a zlib or gzip wrapper encapsulating the data, by adding a header and footer. This provides stream identification and error detection that are not provided by the raw DEFLATE data.

The zlib wrapper (RFC 1950)^[4] is smaller than the gzip wrapper (RFC 1952),^[6] as the latter stores a file name and other file system information.

Algorithm

As of September 2018^[update], zlib only supports one algorithm, called DEFLATE, which uses a combination of a variation of LZ77 (Lempel–Ziv 1977) and Huffman coding.^[7] This algorithm provides good compression on a wide variety of data with minimal use of system resources. This is also the algorithm used in the Zip archive format. The header makes allowance for other algorithms, but none are currently implemented.

Resource use

zlib provides facilities for control of processor and memory use. A compression level value may be supplied that trades speed for compression. There are also facilities for conserving memory, useful in restricted memory environments, such as some embedded systems.

Strategy

The compression can be optimized for specific types of data. If one is using the library to always compress specific types of data, then using a specific strategy may improve compression and performance. For example, if the data contain long lengths of repeated bytes, the run-length encoding (RLE) strategy may give good results at a higher speed. For general data, the default strategy is preferred.

Error handling

Errors in compressed data may be detected and skipped. Further, if "full-flush" points are written to the compressed stream, then corrupt data can be skipped, and the decompression will resynchronize at the next flush point—although no error recovery of the corrupt data is provided. Full-flush points are useful for large data streams on unreliable channels, where some data loss is unimportant, such as in some multimedia applications. However, creating many flush points can affect the speed as well as the amount (ratio) of compression.

Data length

There is no limit to the length of data that can be compressed or decompressed. Repeated calls to the library allow an unlimited number of blocks of data to be handled. Some ancillary code (counters) may suffer from overflow for long data streams, but this does not affect the actual compression or decompression.

When compressing a long (or infinite) data stream, it is advisable to write regular full-flush points.

Forks

A commonly used library built on an old codebase, zlib is also frequently forked by third-parties that claim improvements to this library:

• Intel has a high-performance fork of zlib.^[13]
• CloudFlare maintains a high-performance fork with "massive" improvements.^[14]
• zlib-ng is a zlib replacement fork for "next generation" systems. It removes workaround code for compilers that do not support ANSI C, integrates Cloudflare and Intel optimizations, adds hardware acceleration (SIMD and intrinsic functions), and uses code sanitizers, fuzzing, and code coverage to help find bugs.^[15]