Closed GoogleCodeExporter closed 9 years ago
The out-of-memory was caused by a getStreamBytes(InputStream is) call in the
JpegUtils.traverseJFIF method. It reads all the image-data of the image into a
byte-
array. This array can get too large and an out-of-memory can be thrown.
This byte-array is held onto for a while until the image (data and exif) is
finally
written out.
The idea of the fix is this:
Instead of lugging around a large byte-array, why not hold on to the
input-stream
instead. This will save a lot of memory. When the image-data needs to be
written
out, instead of writing a byte-array, the contents of the input-stream are
written
out instead.
Changes:
Add one more method to JPegUtils.Visitor interface:
public boolean visitSOS(int marker, byte markerBytes[], InputStream is);
It differs from the other 'visitSOS' method in that it accepts an input-stream
instead of a byte-array.
This method returns true only if the implementor of this method will close the
given
InputStream itselft. If the method returns false, the caller should close the
InputStream.
Change all implementations of JpegUtils.Visitor to implement this new method.
All but two of the implementations do absolutely nothing when this method is
called,
except returning false (the caller should close the specified InputStream).
The other two will store the InputStream in an instance-member:
ExifRewriter.analyzeJFIF implements this.
JpegRewriter.analyzeJFIF implements this.
Both these two implement this method by adding a new new JFIFPieceImageData
(markerBytes, is) to the JFIFPieces.pieces list. When the
JFIFPieceImageData.write
(OutputStream) method is called, the stored InputStream will write all its data
out
to the specified OutputStream (and close the InputStream).
Modify JpegUtils.traverseJFIF:
Remove the call to 'getStreamBytes(is)'.
Call 'doClose = !visitor.visitSOS(marker, marketBytes, is)' instead.
Only if 'doClose' is true, close the inputstream 'is' in the finally block.
Original comment by flyingdu...@gmail.com
on 3 Oct 2009 at 4:22
Original issue reported on code.google.com by
flyingdu...@gmail.com
on 3 Oct 2009 at 4:03