Fix object stream parsing

[michaelweiser]

Commit 8cc27b6 broke object stream parsing by resetting the content
cursor PDFParser.charCounter to zero on every invocation. Unfortunately it's unclear from the change what this was fixing at the time but it certainly broke object stream parsing. Reproducer:

$ echo -e "create pdf\ncreate object_stream\nall\nsave /tmp/foo.pdf" | \
	peepdf -i

Without fix:

$ peepdf -j /tmp/foo.pdf
Error: An error has occurred while parsing an indirect object!!

With this change: JSON output as expected (same for other outputs).

$ peepdf -j /tmp/foo.pdf
{
    "peepdf_analysis": {
[...]
           "version": "0.3"
        }
    }
}

Given the object created by above call in local variable content:

<< /Length 131
/N 3
/Type /ObjStm
/Filter /FlateDecode
/First 14 >>
stream
x<9c>M<8c>[...]
endstream

at https://github.com/jbremer/peepdf/blob/11dab2ead9bf344e6aa8cca7f82a17771350f1be/peepdf/PDFCore.py#L7873 self.charCounter points to the end of the dictionary (>>) and is 65. readUntilSymbol() is supposed to advance it to the stream keyword which is three characters away. It first determines the substring to search based on the current value of 65, then resets the counter to zero, finds the symbol three characters into the string and finally sets self.charCounter to 3. The following calls to readSymbol() and colleagues fail silently but advance charCounter beyond the next newline. From there the next call to readUntilSymbol('endstream') tries to extract the actual objectstream content. Because charCounter is still in the middle of the dictionary, it extracts part of that as well and decoding of the object stream eventually fails.

Some poor man's debug prints showcase the issue:

diff --git a/peepdf/PDFCore.py b/peepdf/PDFCore.py
index a61d776..fc9df3b 100644
--- a/peepdf/PDFCore.py
+++ b/peepdf/PDFCore.py
@@ -7870,18 +7870,22 @@ class PDFParser:
                     nonDictContent = content[self.charCounter:]
                     streamFound = re.findall('[>\s]stream', nonDictContent)
                     if streamFound:
+                        print("1 charCounter: %d" % self.charCounter)
                         ret = self.readUntilSymbol(content, 'stream')
                         if ret[0] == -1:
                             return ret
+                        print("2 charCounter: %d, RET: %s" % (self.charCounter, ret[1]))
                         self.readSymbol(content, 'stream', False)
                         self.readUntilEndOfLine(content)
                         self.readSymbol(content, '\r', False)
                         self.readSymbol(content, '\n', False)
+                        print("3 charCounter: %d" % self.charCounter)
                         ret = self.readUntilSymbol(content, 'endstream')
                         if ret[0] == -1:
                             stream = content[self.charCounter:]
                         else:
                             stream = ret[1]
+                            print("4 charCounter: %d, RET: %s" % (self.charCounter, ret[1]))
                             self.readSymbol(content, 'endstream')
                         ret = self.createPDFStream(dictContent, stream)
                         if ret[0] == -1:

Output without the fix:

peepdf -j /tmp/foo.pdf
1 charCounter: 65
2 charCounter: 3, RET: >>

3 charCounter: 15
4 charCounter: 192, RET: /N 3
/Type /ObjStm
/Filter /FlateDecode
/First 14 >>
stream
x<9c>M<8c>[...]

Error: An error has occurred while parsing an indirect object!!

Output with the fix:

peepdf -j /tmp/foo.pdf
1 charCounter: 65
2 charCounter: 68, RET: >>

3 charCounter: 75
4 charCounter: 207, RET: x<9c>M<8c>[...]

[coveralls]

Coverage decreased (-0.3%) to 25.447% when pulling 431e872 on michaelweiser:object-stream into 11dab2e on jbremer:master.

[michaelweiser]

Pushed some more fixes to object-stream-related problems, particularly one containing javascript or sloppy object reference offsets.