PDF page-stream optimizer library? - pdf

Question

Has anyone written a library (or just a program) that optimizes the contents of PDF page streams? I am talking about things like "delete q...Q blocks that have no overall effect", "merge adjacent BT...ET blocks", "track the graphics state and delete operators that set something to the value it already has", maybe even "reorder drawing operations to minimize graphics state changes, when this can be done without changing the appearance of the page". I ain't picky as to implementation language, but open source is very much preferred, as I may need to hack it up for my particular needs.
Here is a small fragment of an example of what I would like done. R's "grid" graphics + its PDF backend generate ridiculous numbers of pointless operations, like this:
1 J 1 j q
Q q
Q q
Q q
Q q
Q q
Q q
Q q
Q q
Q q
BT
0.000 0.000 0.000 rg
/F2 1 Tf 12.00 0.00 -0.00 12.00 168.43 14.40 Tm [(T) 120 (ask)] TJ
ET
Q q
BT
0.000 0.000 0.000 rg
/F2 1 Tf 0.00 12.00 -12.00 0.00 19.42 205.26 Tm
[(Quer) -15 (ies per min) 10 (ute)] TJ
ET
Q q
Q q 23.02 489.60 26.53 0.00 re W n
Q q
Q q 23.02 489.60 26.53 0.00 re W n
Q q
Q q
Q q
[...]
This could be crushed down to just
1 J 1 j
BT
/F2 1 Tf
12 0 0 12 168.43 14.40 Tm [(T) 120 (ask)] TJ
0 12 -12 0 19.42 205.26 Tm [(Quer) -15 (ies per min) 10 (ute)] TJ
ET
and possibly even further with more sophisticated use of the text operators, which I can't do in my head.

Answer 1

The Java "Multivalent Tools" has a "compress" tool that will do this:
http://multivalent.sourceforge.net/Tools/pdf/Compress.html
The Compress tool was removed from the latest Multivalent jar, but you can download an older version from the following location:
http://code.google.com/p/pdfsizeopt/downloads/detail?name=Multivalent20060102.jar&can=2&q=

Answer 2

That looks remarkably like the PDF output of iText's PdfGraphics2D interface, in a worse-than-usual case. The usual case isn't so hot either, but it's not THAT bad.
If I'm right, the answer is still no, but you can write one yourself, as you clearly have no fear of content streams:
ByteBuffer internalBuf = myPdfContentByte.getInternalBuffer();
String newContents = magic( internalBuf.toString() );
internalBuf.reset();
internalBuf.append( newContents );
magic() is a tad nebulous, but writing code to remove "q Q" pairs should be trivial. Yanking clipping regions with nothing inside them (line-line-line W n) shouldn't be all that much harder with a bit of regEx.
Getting rid of the line cap/line join settings (j & J) when they aren't used would be Harder. Ditto with combining text blocks or dumping redundant changes to the fill/stroke colors, font&size etc.
"Sophisticated use of the text operators" is going to start looking like black magic compiler optimization in short order.
And if this does happen to be iText, we'd all appreciate it if you'd share your code. We'll cheerfully accept just about any PdfGraphics2D output clean up, I assure you.