Don't use delta for tiny files

[alexlarsson]

Files that are smaller than the pathname of the delta source (and some extra header bytes) now are copied directly to the delta rather than being delta:ed, because just the delta header would already be larger than the file content.

This happens a lot in e.g. bootc images, because there are lots of tiny ostree metadata objects that have long pathnames, like the *.file-xattrs-link files that are 46 bytes, but have 110 char long pathnames.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 40.59%. Comparing base (13704c8) to head (00093ac).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main      #70       +/-   ##
===========================================
- Coverage   59.51%   40.59%   -18.92%     
===========================================
  Files          10       10               
  Lines        1035     1037        +2     
===========================================
- Hits          616      421      -195     
- Misses        305      545      +240     
+ Partials      114       71       -43     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[gemini-code-assist]

Code Review

This pull request introduces a check to skip delta generation for files smaller than the source path plus a small overhead, preventing inefficient delta sizes. A suggestion was made to replace the magic number used for the overhead calculation with a named constant to improve code clarity.

[djach7]

LGTM

[rmiki-dev]

LGTM! codecov/project is failing, but once the branch gets rebased and pushed, I should be able to merge

[alexlarsson]

I rebased this.

[alexlarsson]

It seems the code coverage test things the code coverage decreased by 19%, but i'm only adding three lines. Something must be going wrong with it.

[kgiusti]

I think I can explain the drop in codecov - the proposed patch contains an optimization that's really very good. Turns out our test coverage now hits the optimization so we're not executing the old codepaths. Good for the tool, but makes codecov sad.

There's nothing wrong with this patch. But we do need to update the test cases to use data that doesn't trigger the optimization.

Here's a quick compare of main branch coverage vs the PR:

on main:

go tool covdata percent -i=/home/kgiusti/work/containers/tar-diff/test/coverage -o=/home/kgiusti/work/containers/tar-diff/test/coverage/integration.out
	github.com/containers/tar-diff/cmd/tar-diff		coverage: 62.7% of statements
	github.com/containers/tar-diff/cmd/tar-patch		coverage: 60.7% of statements
	github.com/containers/tar-diff/pkg/protocol		coverage: 75.0% of statements
	github.com/containers/tar-diff/pkg/tar-diff		coverage: 67.0% of statements
	github.com/containers/tar-diff/pkg/tar-patch		coverage: 67.1% of statements

from pr:

go tool covdata percent -i=/home/kgiusti/work/containers/tar-diff/test/coverage -o=/home/kgiusti/work/containers/tar-diff/test/coverage/integration.out
	github.com/containers/tar-diff/cmd/tar-diff		coverage: 62.7% of statements
	github.com/containers/tar-diff/cmd/tar-patch		coverage: 60.7% of statements
	github.com/containers/tar-diff/pkg/protocol		coverage: 75.0% of statements
	github.com/containers/tar-diff/pkg/tar-diff		coverage: 43.1% of statements
	github.com/containers/tar-diff/pkg/tar-patch		coverage: 24.4% of statements

[djach7]

I just created an issue to update the tests (#76). This PR is good to merge imo, I think the commits just need to be re-signed.

[alexlarsson]

I signed the commit