QSettings::beginWriteArray() produces incorrect output under Linux (bug?)

[JonB]

Qt 5.12.2 as supplied with Ubuntu 19.04, at least.

See the example code at https://doc.qt.io/qt-5/qsettings.html#beginWriteArray. Additionally I have called QSettings.setDefaultFormat(QSettings.Format.IniFormat), but I don't imagine that's relevant***.

The sample output shown as intended is:

The generated keys will have the form

    logins/size
    logins/1/userName
    logins/1/password
    logins/2/userName
    logins/2/password
    logins/3/userName
    logins/3/password

However, I get that output but with \s where it should be /s, e.g. logins\1\userName :(

We are admonished not to go for backslashes instead of forward slashes in our own keys, for good reason....

Could someone kindly verify under Linux at least whether they get same? What about Windows? I worry about reporting this, because then if they fix it the previously-saved arrays in files may not read in? :(

*** EDIT Hmmm... Perhaps that IniFormat is actually what the behaviour is all about? (I'm just using it at present so all platforms save to a file, not Windows registry.) But there is

On Unix, NativeFormat and IniFormat mean the same thing

and there is

QSettings always treats backslash as a special character and provides no API for reading or writing such entries.

though of course the backslashes I am seeing are in QSettings-generated key names, not my values.

Nope: I just tried removing the IniFormat: Linux saved it as "native" .conf file instead of .ini, but they are still all backslashes.

Still scary!?

[JonB]

For right or for wrong, raised https://bugreports.qt.io/browse/QTBUG-81951. Hopefully someone will check whether it's only a documentation issue or if it has any consequences.

[Pl45m4]

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

logins\1\userName

How did you print this? With toString() or is this in your settings file?

3. Do not use slashes ('/' and '\') in section or key names; the backslash character is used to separate sub keys (see below). On windows '' are converted by QSettings to '/', which makes them identical.

(From https://doc.qt.io/qt-5/qsettings.html#section-and-key-syntax)

Maybe its a conversion issue from QSetting together with Linux OS (I would understand, if it was the other way round)
I can test on Win 10 (Qt 5.11.2) and Ubuntu 18.04 (Qt 5.13.2) later.

UPDATE:

Windows (5.11.2):

• ini file
• QSettings::Format::IniFormat set

[logins]
1\userName=userName
1\password=password
2\userName=userName
2\password=password

Linux (Qt 5.13.2):

• same code / project as on Windows
• same, unsatisfying output... :(

[logins]
1\password=password
1\userName=userName
2\password=password
2\userName=userName

[JonB]

@Pl45m4 said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

logins\1\userName

How did you print this? With toString() or is this in your settings file?

This is what ends up in the file (else I wouldn't be reporting an issue!) :)

As per the docs and the example there I quoted, they claim it should be logins/1/userName, but it's logins\1\userName. Honest, guv!

Do not use slashes ('/' and '') in section or key names; the backslash character is used to separate sub keys (see below). On windows '' are converted by QSettings to '/', which makes them identical.

Yes, that's why they I said they admonish us not to use them ourselves. And I don't. In case you're not aware, this is an issue of what QSettings::beginWriteArray() is producing, not me. They are choosing to implement "save array" via multiple keys-worth of output. The docs say they will use '/', which I feel marginally safe about, but in practice at least under Ubuntu they are outputting \, which I'm less safe about :)

This will presumably be either a big in the (intended) code, or incorrect documentation.

Thank you, yes, please do try on whatever platform(s) you have. Armed with your results, I will presumably then have to go raise a bug report....

EDIT I think your latest modification has just crossed with my reply. Then it looks like it's not any better/different under Windows either....

[Pl45m4]

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

Thank you, yes, please do try on whatever platform(s) you have. Armed with your results, I will presumably then have to go raise a bug report....

Additionally I think there is a mistake or typo in the code example :)

QList<Login> logins; // QList named LOGINS
...

QSettings settings;
settings.beginWriteArray("logins");
for (int i = 0; i < logins.size(); ++i) {
    settings.setArrayIndex(i);
    settings.setValue("userName", list.at(i).userName); // LIST.at(i)???
    settings.setValue("password", list.at(i).password); // LIST.at(i)??? Shouldn't it be "logins.at(i)"?!
}

Discovered another weird thing, as I'm comparing the ini file on Window with the ini file on Linux.
When you look at the order, you will see that Linux wrote password, userName, while on WIndows it's userName, password (as supposed due to code, where "userName" is written to settings first)...

[JonB]

@Pl45m4
My code is actually in Python, it's the same as the example but I didn't copy that bit of the C++ code so I didn't notice. Yes, using list in their code example is wrong, it should be logins :)

I wouldn't rely on the order being the same from code to file or across platforms. QSettings probably holds some unordered dictionary for the keys, so output order would not be guaranteed. And that would be reasonable for ini file/registry saving.

The only bit I care about is whether the file ends up with forward or backward slashes!

[Pl45m4]

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

The only bit I care about is whether the file ends up with forward or backward slashes!

The slashes in the doc could be just for demonstration purposes. Just to show the QSetting hierarchy?!

• The generated keys will have the form

What if "form" is meant in terms of hierarchy, separated by single slashes and not showing the actual content of the save file?!

If this is the case, it's very misleading, but occording to QSettings doc (at least on Windows) QSettings ignores '\' and converts them to '/' anyway...

So it could be an unfortunately worded example for this:

- logins
-------- 1
------------- userName
------------- password
-------- 2
------------- userName
------------- password

Edit:

Looks like arrays always have backslashes (if they should? I dont know...)
https://forum.qt.io/topic/79494/qsetting-array/4

Just found a version of QSettings code online.
https://github.com/radekp/qt/blob/master/src/corelib/io/qsettings.cpp

void QSettingsPrivate::processChild(QString key, ChildSpec spec, QMap<QString, QString> &result)
{
    if (spec != AllKeys) {
        int slashPos = key.indexOf(QLatin1Char('/')); // ### SLASH
        if (slashPos == -1) {
            if (spec != ChildKeys)
                return;
        } else {
            if (spec != ChildGroups)
                return;
            key.truncate(slashPos);
        }
    }
    result.insert(key, QString());
}

void QSettingsPrivate::beginGroupOrArray(const QSettingsGroup &group)
{
    groupStack.push(group);
    if (!group.name().isEmpty()) {
        groupPrefix += group.name();
        groupPrefix += QLatin1Char('/'); // # SLASH!!
    }
}

In addition to that, QSettings uses the QDir::separator() which indeed returns a single slash ('/') on my Linux system and a backslash on Windows....

I really don't know where the backslash is coming from and WHY? :-)

[JonB]

@Pl45m4
From your https://forum.qt.io/topic/79494/qsetting-array/4 it looks like it is producing \s elsewhere. At least I know. Then it's perhaps just a docs issue.

[Pl45m4]

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

@Pl45m4
From your https://forum.qt.io/topic/79494/qsetting-array/4 it looks like it is producing \s elsewhere.

Just because others face the same issue or behavior, it doesn't mean that it's intended or correct :)
Could be wrong (buggy) since years / many releases, but seems to work :)

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

Then it's perhaps just a docs issue.

The second part of my post (the Qt source code) shows that it's not a doc issue. You can also find the same comment in the source code ("The generated keys will have the form.... bla...")... Really weird...

[Christian Ehrlicher]

What's the problem as long as beginReadArray() can read them later on?

[JonB]

@Christian-Ehrlicher
At the time I raised the question I did not yet know whether it could be read OK. Now I am pleased to report it does. I merely asked to see if it was the same for others. I do not know if there any other consequences, hopefully not.

So you belong to the "it doesn't matter what the docs say as long as it works" school ;-)

[Pl45m4]

Nevertheless, it would be satisfying (for me as well) if someone could explain where the backslashes come from, because even in current sources arrays start explicitly with '/'. (https://github.com/radekp/qt/blob/master/src/corelib/io/qsettings.cpp#L393)

[JonB]

@Pl45m4

Nevertheless, it would be satisfying (for me as well) if someone could explain where the backslashes come from

All backslashes live in a land of their own, where everything is the wrong way round. They emanate from there, mystically. And sometimes arrive in pairs.

I will raise a report on the Qt bug forum tomorrow and they can decide whether it's a documentation or behaviour issue.

[Christian Ehrlicher]

@JonB said in QSettings::beginWriteArray() produces incorrect output under Linux (bug?):

So you belong to the "it doesn't matter what the docs say as long as it works" school ;-)

No, but to the 'I don't care about the internals as long as the library can read it again'. Or do you really care how e.g. QDataStream encodes and decodes the stuff as long as it is doing what you expect?

[JonB]

@Christian-Ehrlicher
So are you saying you'd rather I didn't report it? When the docs choose to tell me the file will look like such & such and I try it and it does not, I shouldn't question it, shouldn't ask about it here but just assume it's fine? I don't understand your point/suggestion?

[Christian Ehrlicher]

From my pov that's only a documentation error.

[JonB]

@Christian-Ehrlicher
No problem. I'll log it tomorrow, if it's only docs and the behaviour is intended then at least someone (might be you!) will have seen it and can decide.

[JonB]

For right or for wrong, raised https://bugreports.qt.io/browse/QTBUG-81951. Hopefully someone will check whether it's only a documentation issue or if it has any consequences.