apt package repositories/format question

Michael_S · 12-04-2006, 10:10 PM

I notice that every time I do a dist-upgrade, something like this happens: 200 MB of updated packages are downloaded, and after unpacking 5 MB of extra disk space is used.

It seems like if I have package foo version 6 and package foo version 7 is released, the full package file for version 7 is downloaded.

Wouldn't it make more sense for the apt repositories to store each package as a full .deb file, and also as a set of binary patches that allow upgrading previous releases to the newest release?

If you don't have any version of a package you request, apt-get downloads the full .deb file. If you do have some version of the package, apt-get downloads and applies a series of binary patches to that deb file to transform it into the version you want.

Over the long term, I think the bandwidth savings for this, and the speed of distributing upgrades, would be enormous.

pljvaldez · 12-05-2006, 11:51 AM

Interesting thought. Although I have to admit that my only experience with "patch" upgrades is with windows, and it always seems to result in buggy programs after a couple of upgrades. Of course, it's entirely possible that's a Windows problem and not a "patching" problem...

Michael_S · 12-05-2006, 02:40 PM

Well, more explicitly, this is what I have in mind:

1. Repository gets package foo0.5 and its md5sum

2. PC downloads package foo0.5, verifies md5sum, installes package, keeps package .deb file.

3. Repository gets package foo0.6, its md5sum, and package_diff_foo0.5-0.6

4. PC downloads package_diff_foo0.5-0.6 and the md5sum for package foo0.6.

5. PC copies the foo0.5.deb files to foo0.5.deb.tmp and applies the package_diff files to it.

6. PC compares the md5sum of the resulting file to foo0.6. If there is a match, it renames the resulting file to foo0.6. If there is no match, it deletes the corrupt file and downloads the full package from the repository.

7. Now the foo0.6 file on the PC should match the one in the repository exactly.

I know incremental patching is in place for kernel builds. I'm sure incremental binary patches would be more complicated. But I would think this is worthwhile.