[PR #426] [MERGED] Create large window brotli, new incompatible compression format. #1036

New Issue

claunia · 2026-01-29T20:52:13Z

claunia commented

2026-01-29 20:52:13 +00:00

📋 Pull Request Information

Original PR: https://github.com/google/brotli/pull/426
Author: @Zip753
Created: 9/14/2016
Status: ✅ Merged
Merged: 9/19/2016
Merged by: @eustas

Base: large-window ← Head: large-window

📝 Commits (2)

12f30e2 Extend Brotli to use large window (1GB).
7928970 Comply to C99 style.

📊 Changes

15 files changed (+117 additions, -47 deletions)

View changed files

📝 common/constants.h (+4 -2)
📝 dec/decode.c (+14 -4)
📝 enc/backward_references.c (+7 -5)
📝 enc/backward_references.h (+1 -0)
📝 enc/brotli_bit_stream.c (+32 -16)
📝 enc/brotli_bit_stream.h (+1 -0)
📝 enc/command.h (+19 -9)
📝 enc/encode.c (+13 -2)
📝 enc/metablock.c (+9 -2)
📝 enc/metablock.h (+1 -0)
📝 enc/prefix.h (+4 -2)
📝 enc/quality.h (+3 -1)
📝 enc/ringbuffer.h (+6 -2)
📝 include/brotli/encode.h (+2 -1)
📝 tools/bro.c (+1 -1)

📄 Description

Large window brotli is a brotli version that can use 30-bit (1GB) window. It uses the reserved short-code 0010001 to indicate 30-bit window, therefore breaking section 9.1 of RFC 7932.

This is an experimental change, so that brotli can be more competitive on large inputs.

The compression ratio on enwik9 for quality 11 went from 4.47 for 24-bit window to 5.02 for 30-bit window. Also, on 10GB corpus the compression ratio for quality 10 and 11 increased from 2.52 and 2.54 to 2.78 and 2.80 respectively.

Additionally, large window brotli has been very useful in researching brotli distance encoding and exploring patching possibilities.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/google/brotli/pull/426 **Author:** [@Zip753](https://github.com/Zip753) **Created:** 9/14/2016 **Status:** ✅ Merged **Merged:** 9/19/2016 **Merged by:** [@eustas](https://github.com/eustas) **Base:** `large-window` ← **Head:** `large-window` --- ### 📝 Commits (2) - [`12f30e2`](https://github.com/google/brotli/commit/12f30e224ac2982144a90b611691161009a2a574) Extend Brotli to use large window (1GB). - [`7928970`](https://github.com/google/brotli/commit/792897037ce100ed6a19cc59894b70b45ce9bac5) Comply to C99 style. ### 📊 Changes **15 files changed** (+117 additions, -47 deletions) <details> <summary>View changed files</summary> 📝 `common/constants.h` (+4 -2) 📝 `dec/decode.c` (+14 -4) 📝 `enc/backward_references.c` (+7 -5) 📝 `enc/backward_references.h` (+1 -0) 📝 `enc/brotli_bit_stream.c` (+32 -16) 📝 `enc/brotli_bit_stream.h` (+1 -0) 📝 `enc/command.h` (+19 -9) 📝 `enc/encode.c` (+13 -2) 📝 `enc/metablock.c` (+9 -2) 📝 `enc/metablock.h` (+1 -0) 📝 `enc/prefix.h` (+4 -2) 📝 `enc/quality.h` (+3 -1) 📝 `enc/ringbuffer.h` (+6 -2) 📝 `include/brotli/encode.h` (+2 -1) 📝 `tools/bro.c` (+1 -1) </details> ### 📄 Description Large window brotli is a brotli version that can use 30-bit (1GB) window. It uses the reserved short-code `0010001` to indicate 30-bit window, therefore breaking [section 9.1 of RFC 7932](https://tools.ietf.org/html/rfc7932#section-9.1). This is an experimental change, so that brotli can be more competitive on large inputs. The compression ratio on enwik9 for quality 11 went from 4.47 for 24-bit window to 5.02 for 30-bit window. Also, on [10GB corpus](http://mattmahoney.net/dc/10gb.html) the compression ratio for quality 10 and 11 increased from 2.52 and 2.54 to 2.78 and 2.80 respectively. Additionally, large window brotli has been very useful in researching brotli distance encoding and exploring patching possibilities. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

claunia added the pull-request label 2026-01-29 20:52:13 +00:00

claunia referenced this issue

2026-01-29 20:54:47 +00:00

[PR #1036] [MERGED] add option to delete files that are not "compressed" #1580

claunia referenced this issue

2026-01-29 20:54:48 +00:00

[PR #1036] add option to delete files that are not "compressed" #1585

Sign in to join this conversation.