Streaming custom QVariant?

[wysota]

So, for example, when reading a binary file of various custom types, how would one inspect what the record actually is? Qt's serialization would have to support this. Something has to handle the data as a generically sized "block" for the streaming to work. Otherwise, wouldn't know how to iterate through the binary file.

QDataStream is a low level mechanism. You want a high level one thus you have to use this low level mechanism to implement a high level mechanism. Simply speaking, if you want to detect what is the type of data in the stream, stream additional data that will let you detect the type. You can stream the name of the type, for example.

QDataStream stream(...);
QString str;
stream << "string" << str;
...
QDataStream inStream(...);
QString type;
stream >> type;
QVariant var;
if(type=="string"){
  QString s;
  stream >> s;
  var.setValue(s);
}

QDataStream stream(...);
QString str;
stream << "string" << str;
...
QDataStream inStream(...);
QString type;
stream >> type;
QVariant var;
if(type=="string"){
  QString s;
  stream >> s;
  var.setValue(s);
}

To copy to clipboard, switch view to plain text mode 

You can use QMetaType to have automatic coversion of the type id to the name of the type to avoid unnecessary comparisons like in my code above.

BTW, I realize that QVariant (on its own) can't recreate every possible structure in the world, but through template-programming was thinking it should be able reconstruct any type (from the type identifier in the header).

I'm not sure how templates would be helpful here. QVariant is a class on its own, you don't create a new "QVariant" class in each application that uses Q_DECLARE_METATYPE.

The documentation (that I've come across) is limited on serialization. Do you have anything you'd recommend? Seems like it would be a great tutorial (maybe it is).

The only relevant docs are in QDataStream reference. This is really a marginal topic in Qt.

[jpn]

See [WIKI]Using custom data types with Qt[/WIKI].

[brcain]

QDataStream is a low level mechanism. You want a high level one thus you have to use this low level mechanism to implement a high level mechanism. Simply speaking, if you want to detect what is the type of data in the stream, stream additional data that will let you detect the type. You can stream the name of the type, for example.

Thanks again. And I hope I don't sound too critical/whiney ...

This seems like standard serialization framework support that should already exist. Maybe I misread what serialization support is actually there. The reason for serialization is so that you don't have to handle the meta-type information that is wrapped around the user type ... but are rather given the meta-type information that you can cast (from some data buffer) to what actually type is.

Shouldn't serialization framework automatically handle Qt types and provide mechanism for custom types? I do override the QDataStream streaming operators, but that doesn't say anything about the header (type, size, etc.) that should be handled automatically.

I can punt on this / quit beating a dead horse ... am just very surprised.

[brcain]

So, how do you stream from a binary file of varying types in non-deterministic order using QDataStream?

Can you even do this with pre-defined Qt types?

[wysota]

Shouldn't serialization framework automatically handle Qt types and provide mechanism for custom types?

It does. But not through QVariant

I do override the QDataStream streaming operators, but that doesn't say anything about the header (type, size, etc.) that should be handled automatically.

That's your task in the implementation of the operator.

QDataStream is a mechanism do stream binary data to/from a device. It is not a general-purpose serialization mechanism but it can be used to create one. It's main task is to provide a way to store (and retrieve) binary data in some well known internal data format where you know what data you expect.

[brcain]

QDataStream [...] is not a general-purpose serialization mechanism but it can be used to create one.

Okay, I think I finally get it. I read too much into the serialization support.

A serialization framework would be a great future capability. It could be used for binary file streams as well as network streams, etc. ... as a tidey way to packetize all records.

Much thanks,
Ben