Question

I am looking at compressing some very large strings (text fields) in Ruby before inserting them into database blob fields. Compression by itself is easy, I can just use Zlib.

However, I am also looking at instances when I may have similar copies of strings. Eg. I might have something already stored in the database – stringA. A modification gives me stringB. I want to store a compressed version of the difference between stringA and stringB, so that if I have stringA and the compressed diff, I can get stringB back.

Is there a suitable library for this?

Ideally, it would be a single step binary diff compression. I don’t really want a human-readable text diff (which might waste more space). It only needs to be machine readable. Therefore please don’t suggest that I compress using diff -u oldFile newFile > mods.diff and patch < mods.diff.

Answer

Edit: Thank you Mark Adler for part of the answer (didn’t know there was a set_dictionary method). I want to do this in Ruby, thus, the relevant method name is set_dictionary. However, trying to get this done is much more difficult than without the dictionary.

Without using a dictionary, we can do:

A = "My super string to be compressed. Compress me now to " \
    "save the space used to store this super string."
cA = Zlib::Deflate.deflate(A)
# => "x\234U\214\301\r\200 \020\004[\331\nh\302\267E n\224\a\034\271;4v..."

Zlib::Inflate.inflate(cA)
# => "My super string to be compressed. Compress me now to save the..."

But to use a dictionary, you need to make sure to pass Zlib::FINISH to deflate to flush the output, and allow a Zlib::NeedDict exception before adding the dictionary when inflating:

B = "A super string with differences, let's see how much " \
    "extra space the differences will take in this super string!"
zlib_deflate = Zlib::Deflate.new
zlib_deflate .set_dictionary(A)
dB = zlib_deflate .deflate(B, Zlib::FINISH)
# => "x\2733\324$\230sD\265\242<\263$C!%3--\265(5/9\265XG!'\265D\035\250..."

zlib_inflate = Zlib::Inflate.new
zlib_inflate.inflate(dB) # Exception thrown
# => Exception: Zlib::NeedDict: need dictionary
zlib_inflate.set_dictionary(A)
zlib_inflate.inflate(dB)
# => "A super string with differences, let's see how much extra space the..."

Answer 1

You can do this with zlib as well. Use the deflateSetDictionary() function to provide stringA as the dictionary when compressing stringB. On the other end you already have stringA when decompressing stringB, so use inflateSetDictonary() with stringA before decompressing stringB.

zlib will then find parts of stringB that match stringA and point to those parts in stringA.

You can do better still by providing stringA and stringB concatenated as the dictionary when compressing stringC. And so on. The dictionary can be up to 32K bytes.