Description
We are using Extractcode to extract large open source archives.
We have found a case where after extraction files some files from subdirectories are extracted into the extraction root and given very long and partially gibberish names.
This happened specifically upon extracting qt6-webengine 6.8.2+dfsg-4 (http://deb.debian.org/debian/pool/main/q/qt6-webengine/qt6-webengine_6.8.2+dfsg-4.debian.tar.xz).
After extraction the file qt6-webengine-6.8.2+dfsg/LICENSES/Qt-GPL-exception-1.0.txt was found in the extraction root with filename qt6-webengine-6.8.2Min _nyopeQu _ GPL-exception-1.0.txt
After some investigation we found that +dfsg/LICENSES/Qt- is being incorrectly decoded as UTF-7, leading to a full-path of qt6-webengine-6.8.2痻\u20fc눂ე葋\uf42dGPL-exception-1.0.txt which is the transliterated to qt6-webengine-6.8.2Min _nyopeQu _ GPL-exception-1.0.txt This means that the directory structure of the file is entirely lost and the file gets extracted to the extraction root instead of the correct directory.
The same happens for a selection of other files in the archive, but not reliably for every file. We were not able to determine what specifically triggers the issue, our assumption is that it has to do with Kanji encoding in some form (e.g. if the final part of a path can theoretically be decoded as a Kanji it incorrectly assumes it is UTF-7 encoded).
We have attached a minimal example of this (note: we had to package it as a .zip due to file support limitations on github, for testing it should be extracted and then reachived as a .tar.xz) with only this Qt-GPL-exception-1.0.txt file contained for easier verification and testing.
qt6-webengine-6.8.2+dfsg.zip
We were able to reproduce the issue reliably on a Debian 13 system, the issue was not reproducible on a Windows 11 machine.
Activity
Anurag-S1ngh commentedon Oct 31, 2026
Decode archive paths as UTF-8 first aboutcode-org#77
srehm commentedon Nov 11, 2026
The root cause in in chardet which is used (if installed) by bs4 UnicodeDammit which is in turn used by commoncode.
I have opened an issue there: chardet/chardet#432 (comment)
But maybe the issue calls for a more general solution: the problem with all charset detectors is that they fail quite often when the input is as small as in our case.
In the meantime, defaulting to utf-8 as suggested by shashank-tomar0 is not the worst idea IMHO.