git diff generation is sometimes wrong

[Robie Basak]

https://code.launchpad.net/~ahasenack/ubuntu/+source/apache2/+git/apache2/+merge/348679 is an example of where the generated diff seems to include far more than just what changed.

17:02 <nacc> ahasenack: rbasak: yeah, i dug into it and it seems like it's something was in pygit2 or libgit2, but i forget immeidately

17:08 <nacc> ahasenack: in theory, we should be able to reproduce the diff that LP shows via turnip (iirc, that's the name of the project)

17:08 <nacc> ahasenack: by tracing that code to the diff generation (which I believe is pretty straightforward) and then seeing if we can get the same thing in a simple tool

17:11 <cjwatson> Start from turnip.api.views.DiffMergeAPI

17:12 <cjwatson> And if you want to reproduce exact versions and such, use a trusty container with ppa:launchpad/ubuntu/ppa

Filing to keep this useful information in one place, but I think it'll be a while before I investigate.

[Andreas Hasenack (ahasenack)]

https://code.launchpad.net/~paelzer/ubuntu/+source/qemu/+git/qemu/+merge/348672 is another example. Launchpad is showing a diff over a million lines long, when in fact it's much smaller:

(fix-1587065-guest-agent-double-freeze)andreas@nsnx:~/git/packages/qemu/qemu$ git diff pkg/ubuntu/xenial-devel|diffstat
 changelog | 8 +
 patches/series | 1
 patches/ubuntu/lp-1587065-qga-ignore-EBUSY-when-freezing-a-filesystem.patch | 44 ++++++++++
 3 files changed, 53 insertions(+)