when LANG="" file with "ö" in name cannot be opened

[ChristianS99]

Overview

When LANG is set to "" file "Passwörter.kdbx" cannot be opened

Steps to Reproduce

$ keepassxc-cli ls Passwörter.kdbx 
Passwort zum Entsperren von Passwörter.kdbx eingeben:
<...>
$ LANG= keepassxc-cli ls Passwörter.kdbx                                                           
Failed to open database file Passw?rter.kdbx: not found

Expected Behavior

file can be opened

Context

Not 100% sure, but to my experience, LANG should only affect the used output lanugage, and should not affect the encoding that is used for interpreting given arguments, eg.:

$ cat föo                                                                                                                                                                                                               
dasd                                                                                                                                                                                                                                        
$ LANG= cat föo                                                                                                                                                                                                         
dasd

$ LANG= keepassxc-cli --debug-info
KeePassXC - Version 2.7.9
Revision: 8f6dd13

Qt 5.15.14
Debugging mode is disabled.

Operating system: Gentoo Linux
CPU architecture: x86_64
Kernel: linux 6.10.3-gentoo-dist

Enabled extensions:
- Browser Integration
- Passkeys
- SSH Agent
- KeeShare
- YubiKey
- Secret Service Integration

Cryptographic libraries:
- Botan 3.2.0

GUI is affected in same way

[droidmonkey]

This is definitely a qt issue. There isn't much we can do about this. Obviously easy fix is to not use non-ascii characters in your file name or make sure your LANG is set.

[phoerious]

Turns out, this is a Qt issue, but the problem is in fact a bit more complex.

"Passw?rter.kdbx" means a Latin1 string is being interpreted as UTF-8 here. However, simply running

LANG= keepassxc-cli ls Passwörter.kdbx

isn't a good problem demonstration. keepassxc-cli will use a default locale, but the parameters are in whatever your terminal's own input encoding is, so this could be anything.

The following Python snippet is a more stable test case:

import subprocess
subprocess.Popen([b'keepassxc-cli', b'ls', 'Passwörter.kdbx'.encode('iso-8859-1')],
                  env={'LANG': ''}).wait()

This is where we parse the command line arguments: https://github.com/keepassxreboot/keepassxc/blob/develop/src/cli/keepassxc-cli.cpp#L194

I believe that

    for (int i = 0; i < argc; ++i) {
        arguments << QString(argv[i]);
    }

is indeed wrong. This should at least be QString::fromLocal8Bit(argv[i]), but according to the docs, this is equivalent to QString::fromUtf8() on Linux, which is obviously wrong when your system locale isn't Unicode-based.

I think defaulting to UTF-8 is a sane assumption for most Linux systems, but if your input encoding is something else, this will obviously fail. To fix this, we'd need to parse LANG or LC_ALL ourselves, but even with those variables set, we could only guess what the actual input encoding of the command line parameters is.

[ChristianS99]

Turns out, this is a Qt issue, but the problem is in fact a bit more complex.

"Passw?rter.kdbx" means a Latin1 string is being interpreted as UTF-8 here. However, simply running

yeah, agree. looks, like output is latin1, where the terminal expects utf8

LANG= keepassxc-cli ls Passwörter.kdbx

isn't a good problem demonstration. keepassxc-cli will use a default locale, but the parameters are in whatever your terminal's own input encoding is, so this could be anything.

terminal's input encoding is utf8, even with LANG="" as setting this in front of command only change it for the command, not the terminal

The following Python snippet is a more stable test case:

import subprocess
subprocess.Popen([b'keepassxc-cli', b'ls', 'Passwörter.kdbx'.encode('iso-8859-1')],
                  env={'LANG': ''}).wait()

This is where we parse the command line arguments: https://github.com/keepassxreboot/keepassxc/blob/develop/src/cli/keepassxc-cli.cpp#L194

I believe that

    for (int i = 0; i < argc; ++i) {
        arguments << QString(argv[i]);
    }

is indeed wrong. This should at least be QString::fromLocal8Bit(argv[i]), but according to the docs, this is equivalent to QString::fromUtf8() on Linux, which is obviously wrong when your system locale isn't Unicode-based.

mind, this is the qt6 docs, qt5 is different, and actually fromLocal8Bit and fromUtf8 do different things.

I think defaulting to UTF-8 is a sane assumption for most Linux systems, but if your input encoding is something else, this will obviously fail. To fix this, we'd need to parse LANG or LC_ALL ourselves, but even with those variables set, we could only guess what the actual input encoding of the command line parameters is.

#include <QString>
#include <QTextStream>
#include <iostream>

int main(int argc, char *argv[])
{
    if (argc > 1) {
        QTextStream out(stdout);
        std::cout << argv[1] << std::endl;
        int p = 0;
        while (argv[1][p] != 0) {
            printf("%x ", (unsigned char)argv[1][p]);
            p+=1;
        }
        printf("\n");
        QString s1 = QString(argv[1]);
        out << s1 << Qt::endl;
        QString s2 = QString::fromLocal8Bit(argv[1]);
        out << s2 << Qt::endl;
        QString s3 = QString::fromUtf8(argv[1]);
        out << s3 << Qt::endl;
    }
}

small test program to try a few things. running this program with LANG= ./qttest aäböc gives this output:

aäböc
61 c3 a4 62 c3 b6 63 
a?b?c
a??b??c
a?b?c

line 1: terminal is consistent with encoding of input given and output expected line 2: the encoding actualy is utf8 line3: Qstring obviously converts the byte sequence somehow. line 4 and 5: fromLocal8Bit and fromUtf8 are different (on qt5)

[phoerious]

yeah, agree. looks, like output is latin1, where the terminal expects utf8

This is not just the terminal output, but first and foremost the file name. File names are always UTF-8 on Linux, so using a Latin1 string is wrong in any case.

When I QDebug my QLocale, it always says "Latin1", even when it's actually UTF-8. I also couldn't find any difference in behaviour between QString(argv[i]) and QString::fromLocal8Bit(argv[i]). However, looking at the Qt source code for QCommandlineParser::process(&QCoreApplication), I figure that QString::fromLocal8Bit(argv[i]) is indeed the correct way.

[AugustoMagalhaes]

Christian, try to execute the executable like this

LANG=de_DE.UTF-8 executable args

Tell me if it works, I had the same problem in Qt and fixed it by using LANG=C in another situation. Good luck mate

[ChristianS99]

LANG=de_DE.UTF-8 executable args

This works, and it is my default. It just stumbled over the problem by accident, and thought I ccould report it.