Custom image plugin testing difficulties

[SGaist]

The most simple is to copy your plugin in Qt's plugin folder or you can tell your application to search in the folder containing your plugin with QCoreApplication::addLibraryPath

[ChrisM87]

Well as I showed above, with the QT_DEBUG_PLUGINS variable set, the program is picking it up. However, it still refuses to load it. Is the IID or keys array invalid? I figured if it was choking on the plugin code itself, it would either error, segfault, or spit out a corrupt PNG.

[SGaist]

Without seeing your code, it's pretty much Crystal Ball Debugging (™) I can't say

[ChrisM87]

Well, I had posted my code in the original post. However, my handler class was tweaked a bit after spotting some errors, so here's the updated class code:

EDIT: I did a few tests on QImageReader that should fail with unsupported format error, but I keep getting "unknown error." I attempted to load my ".pro" qmake file to see which error it would get but still unknown error. However, I compiled qtraw from Github, dropped it into an imageformats subdirectory and it ran fine.

#include <QDataStream>
#include <QImage>
#include <QIODevice>
#include <QtGlobal>
#include <QtMath>
#include <QVariant>
#include <QVector>
#include <QImageIOHandler>

class TimHandler final: public QImageIOHandler
{
public:
	bool canRead() const;
	bool read(QImage *img);
	QVariant option(ImageOption op) const override;
	bool supportsOption(ImageOption op) const override;
};

#define BIT16 0.121568627451

bool TimHandler::canRead() const
{
	return device()->peek(4) == "\x10\x0\x0\x0";
}

static inline quint32 rgba5551ToArgb32(quint16 c)
{
	quint32 r = ((c & 0xf800) >> 11) / BIT16;
	quint32 g = ((c & 0x7c0) >> 6) / BIT16;
	quint32 b = ((c & 0x3e) >> 1) / BIT16;
	quint32 a = c & 1 ? 0 : 0xff;
	
	return ((a << 24) | (r << 16) | (g << 8) | b);
}

bool TimHandler::read(QImage *img)
{
	QDataStream ds(device());
	ds.setByteOrder(QDataStream::LittleEndian);
	quint32 magic, flags;
	
	ds >> magic >> flags;
	if (ds.status() != QDataStream::Ok || magic != 0x10)
		return false;
	bool hasClut = flags & 0x8 ? true : false;
	quint8 pxlMode = flags & 0x7;
	
	quint8 bpp;
	switch (pxlMode)
	{
		case 0: bpp = 4; break;
		case 1: bpp = 8; break;
		case 2: bpp = 16; break;
		case 3: bpp = 24; break;
		default: bpp = 4;
	}
	
	QVector<QRgb> clut;
	if (hasClut)
	{
		quint32 colors, npals;
		ds.skipRawData(12);
		ds >> colors >> npals;
		
		for (quint16 i = 0; i < (colors*npals); i++)
		{
			quint16 c;
			ds >> c;
			clut.append(rgba5551ToArgb32(c));
		}
	}
	
	ds.skipRawData(8);
	quint16 w, h;
	ds >> w >> h;
	w = (w*16)/bpp;
	
	if (img) delete img;
	img = new QImage(w, h, QImage::Format_ARGB32);
	if (hasClut) img->setColorTable(clut);
	
	for (quint16 y = 0; y < h; y++)
	{
		for (quint16 x = 0; x < w; x++)
		{
			switch (bpp)
			{
				case 4:
				{
					quint8 p;
					ds >> p;
					img->setPixel(x++, y, p & 0xf0);
					img->setPixel(x, y, p & 0xf);
				}
					break;
				case 8:
				{
					quint8 p;
					ds >> p;
					img->setPixel(x, y, p);
				}
					break;
				case 16:
				{
					quint16 p;
					ds >> p;
					img->setPixel(x, y, rgba5551ToArgb32(p));
				}
					break;
			}
		}
	}
	
	return true;
}

bool TimHandler::supportsOption(ImageOption op) const
{
	switch (op)
	{
		case Size:
			return true;
		default:
			return false;
	}
}

QVariant TimHandler::option(ImageOption op) const
{
	switch (op)
	{
		case Size:
		{
			quint32 size;
			quint16 w, h;
			QDataStream ds(device());
			ds.skipRawData(8);
			ds >> size;
			size -= 4; // 12 - image size - palette x - palette y
			ds.skipRawData(size);
			ds >> w >> h;
			return QSize(w, h);
		}
		default:
			return QVariant();
	}
}

[SGaist]

Sorry about the code part, I had a problem when I read it first and only saw the end of it.
What about a more simple json ?

{
    "Keys": [ "tim" ],
    "MimeTypes": [ "image/x-tim" ]
}

[ChrisM87]

Well, while I could try that, I don't know if that's the mime type for TIM files. A Google search doesn't help either. My examination of qtraw shows that it too lacks a mime type entry. Thus, I thought it was unnecessary.

EDIT: tried it anyway, still nothing

[SGaist]

Just thought of something… Your plugin code is correct but:

img.load("0025.tim", "tim");

"0025.tim" is a relative path, is that image in the same folder as your executable ? You should rather give the full path to it.

[ChrisM87]

Same directory, yep. Though of course in a real world scenario this would be different.

[SGaist]

I forgot a correction I made:

Q_PLUGIN_METADATA(IID "org.qt-project.Qt.QImageIOHandlerFactoryInterface" FILE "tim.json")

[ChrisM87]

Yeah, I changed it to that in my progress trying to fix it, but no go. I booted up QtCreator and created a new project from Qt plugin template but I see nothing unusual in comparison with my code, aside from the constructor with QObject* parameter, which I skipped in my code because I saw no point for an empty constructor/destructor.

[ChrisM87]

So after a little tinkering, I got qmake to generate main.moc which I thought was needed, but I'm still having issues with the plugin not being discovered. Additionally, the RGBA5551 conversion function was incorrect, which I fixed, but that shouldn't be the make-or-break factor.

[SGaist]

Not being discovered or not loaded ? Did you put your plugin in a imageformats subfolder ?

[ChrisM87]

Yep. Even set target.path in my .pro file thinking that was the issue. Qt seems very nitpicky.

EDIT: Just debugged again, found this gem of an error:

QLibraryPrivate::loadPlugin failed on "/home/admin/Documents/Qt-extra/imageformats/qtim/timtest/imageformats/libqtim.so" : "Cannot load library /home/admin/Documents/Qt-extra/imageformats/qtim/timtest/imageformats/libqtim.so: (/home/admin/Documents/Qt-extra/imageformats/qtim/timtest/imageformats/libqtim.so: undefined symbol: _ZN11QTimHandler9setOptionEN15QImageIOHandler11ImageOptionERK8QVariant)"

[SGaist]

Which version of Qt are you using ?
What happens if you try to load it by hand using QLibrary ?