Source Caption Decompiler

This is a simple utility to decompile caption .dat files in the Source engine.

Usage

1. Place closecaption_languagename.dat (the file you want to decompile) alongside decompile.py
2. Run python decompile.py closecaption_languagename.dat

There are also some optional commandline arguments that can be used:

• filename or -i filename determines the compiled source caption .dat file. If unspecified, this will be closecaption_english.dat.
• -o filename determines the target output decompiled caption .txt file. If unspecified, this will be closecaption_decompiled.txt.
• -l filename1 filename2... determines the caption names lists to use. If unspecified, this will default to ./lists/tf2.txt ./lists/commentary.txt ./lists/common_cc_emit.txt.
• --nohashsuffix if used, will skip calculating the hash suffix for missing labels. This option is much faster when missing labels are encountered, but means that recompiling the decompiled file won't work for those missing labels.
• --stdout will print the decompiled file into stdout instead of writing a file.
• --raw will ignore proper formatting for decompiled captions files and will only print each label and the caption associated with it.
• -v or --verbose will print out information associated with the decompilation process, intended for use with debugging or interpreting the data of the compiled file.
• -p or --progress will print the current progress of the decompilation process.

Note: if two different caption names end up sharing the same crc32 hash, decompile.py will report that there's been a collision. However, the program will continue regardless -- the decompiled names just may not reflect their contents in the case of a collision.

How it works

The format of compiled captions files looks something like:

- metadata
- list of crc32 hashes of (lowercase) sound names, paired with offsets for the caption strings they reference
- blocks of data that contain the caption strings refered to by the hashes

Even though (by the nature of how hashing works) we can't reverse the crc32 hashes to get the human-readable name directly, we can take lists of human-readable names and generate their hashes to compare against the hashes we find when decompiling. We use lists oriented towards Team Fortress 2, though these lists can be modified to support other Source games.

Unfortunately, it's pretty common for caption files to include things that aren't included in our list. The reasons can range from genuine mistakes on the caption's part (things like including Demoman.PainSevere07 despite not being a valid sound name referenced ingame anywhere) to including custom captions for use in scripts with cc_emit. In these cases we'll decompile it to U.XXXXXXXX.YYYYYYY where XXXXXXXX is the hash written in hex, and YYYYYYY is arbitrary text to ensure that the crc32 hash is the same (so that the decompiled file can be recompiled directly without edits).