Linux shell script to reduce OpenDocument file sizes

[9point9]

In this previous thread I discussed some methods of reducing the size of OpenDocument files losslessly:
http://www.oooforum.org/forum/viewtopic.phtml?t=27339

This is a Linux shell script I have written to automate this:

[AndrewZ]

Though images usually account for most space, you could reduce some space by compacting the XML For example, you could remove all remove redundant namespace declarations and unnecesary whitespace.

[9point9]

[AndrewZ]

[9point9]

I thought you meant editing the files, that would be too complicated!

I've had a go with an xml tool called xmllint. It also uses --nsclean but does'nt use --noindent, I think that is automatic. It is only useful on content.xml, the other files would be the same size if it wasn't for the extra carriage return that it puts at the end of the file.

Using it just on content.xml (and not getting rid of the extra carriage return) and sticking it in my previous script brings the file size down to 677587 from 677642 before.
_________________
Arch Linux
OOo 3.2.0

OOoSVN, change control for OOo documents:
http://sourceforge.net/projects/ooosvn/

[AndrewZ]

[9point9]

I've made quite a few changes to the script:

[AndrewZ]

I do like your script: I plan to try the JPEG optimizations (nice idea) against one of my web sites. But I have two more thoughts about your script:

First, there is an option in OpenOffice.org for optimizing XML, and it seems to be on by default.

Second, the script may cause data loss in some situations, and this problem could be avoided by checking that the tools exist and that no error has occured. For example, in the following code:

[9point9]

[9point9]

I've made a whole load of minor changes which may reduce file sizes more:

- Used the '-reduce' switch with pngcrush. This losslessly reduces the bit depth of an image whose content could be expressed using a smaller pallette. This will rarely be applied.

- Used the '-rem alla' pngcrush switch. This will remove all metadata from the PNG images. As the metadata of the document is important, not the image, most people don't set the meta-data and there is no real use for it in a document I think it can be safely taken out to save space.

- Used the '--strip-all' jpegoptim switch. This also removes metadata tags to save space, again, I don't think they're that useful.

- Removed the bit about optimising thumbnails as they get deleted later on anyway. If you want them back, my older code does it.

- Got rid of the sed bit. It didn't do anything as xmllint did the same and there's still an empty line at the end.

Running this on the original test file brings it down to 674540 bytes. That's 22.32% compaction. This is only on a particular file, on one of my own I have managed to save 41.45%! I'd be keen to know if any one can beat this!

[9point9]

I've changed the script to use OptiPNG instead of PNGcrush. It currently uses 1080 runs. This is more than is recommended but gives 3Kb better than with only 240 runs. It now takes a lot longer but does get the files smaller.

I hope to incorporate brute-forcing of zlib window size. That should make a factor of 14 slower but there has got to be a reason to have an Athlon 64 FX-60.

OptiPNG is better than PNGcrush so you could always reduce the number of runs and end up with something faster.

This was doen with OptiPNG 0.5 which doesn't seem to include everything that PNGCrush does yet. I would expect it to get better with later versions.

This can reduce the setup guide draft 1 used above to 669518. That's 22.9%.

[9point9]

Implementing advancedCOMP gives some improvement: