This document summarizes a research paper that empirically evaluated how well 11 recommended OSGi best practices for modular dependencies were followed for 372 OSGi bundles. The study found that while some practices like preferring package-level dependencies and using versions were followed over 50% of the time, other practices like avoiding dynamic package imports were not widely adopted. Adhering to the best practices was shown to reduce classpath size by up to 23% and improve performance by up to 13%. The findings suggest that following OSGi best practices can help achieve the goals of modularity and efficiency.

1. Make Software Modular!
And… Efficient?s
An Empirical Evaluation of OSGi Dependencies Best Practices
Lina Ochoa, Thomas Degueule & Jurgen Vinju
Mining Software Repositories (MSR) 2018
CWI - S.W.A.T.

2. “Engineers are very pragmatic; they
may admire beauty, but they seek
utility. I tried to find an explanation
of why “beauty” was useful.”
- David L. Parnas
2

3. Modularity in software
Non-modular software Modular software
3

4. How modularity works in Java
language?
4

5. Modularity in Java + OSGi
Bundle-ManifestVersion: 2
Bundle-Name: Lib1
Bundle-SymbolicName: a.lib1
Bundle-Version: 1.2.2.release-2
Bundle-RequiredExecutionEnvironment: JavaSE-1.8
Export-Package: org.package1;version=“0.0.1”
Import-Package: org.package3;version=“[1.1,1.5]”
Require-Bundle: a.lib3;bundle-version=“2.1.1”
DynamicImport-Package: org.test1
org.package1
org.package2
A B
C D E
lib.jar [Java]
MANIFEST.MF [OSGi]
5
Bundle
28 headers

6. Modularity in Java + OSGi
pkg.students.api
pkg.students.impl
6
bund.math
pkg.math.api
pkg.math.impl
bund.studentsbund.sort
pkg.sort.api
pkg.sort.impl Require-Bundle Import-Package
OSGi document:
450 pages!

7. Which OSGi best practices are
advised?
7

8. Systematic review
Google
search
Select based on
keyword
occurrences
Remove
duplicates
Select based
on relevance
268 156 87 21
((good OR bad OR best) AND (practices OR design)) OR smell
8
OSGi-endorsed
web resources

9. OSGi best practices
B1: Prefer package-level dependencies
B2: Use versions when possible
B3: Export only needed packages
B4: Minimize dependencies
B5: Import all needed packages
B6: Avoid dynamic package imports
B7: Separate implementation, API, and OSGi-
specific packages
B8: Use semantic versioning
B9: Avoid splitting packages
B10: Declare additional dependencies in the
bundle classpath
B11: Import exported API packages
9

10. B1: Prefer package-level dependencies
10
pkg.students.api
pkg.students.impl
bund.othersort
pkg.sort.api
pkg.sort.otherimpl
bund.studentsbund.sort
pkg.sort.api
pkg.sort.impl Import-PackageRequire-BundleImport-Package

11. Are OSGi best practices being
followed?
11

12. Best practices in practice
B1
• 58% of the bundles use Require-Bundle.
B2
• 85% Require-Bundle, 59% Import-Package, and 25%
Export-Package dependencies are versioned.
B3
• 15% of exported packages are imported through
Import-Package.
12
372 bundles

13. Do OSGi best practices have an
impact on the system efficiency?
13

14. Studied corpus
14
Transform
Compare
Benchmark
C0
C6
C1
R6
R1
ResultsCorpora
…
…
372 bundles
Rascal MPL

15. Is classpath size impacted?
15
15.4% 23.3%

16. Is performance impacted?
16
-13.3%
12.8%

17. “I tried to find an explanation of
why “beauty” was useful.”
- David L. Parnas
17

18. Make Software Modular! And… Efficient?
An Empirical Evaluation of OSGi Dependencies Best Practices (MSR 2018)
Lina Ochoa
lina.ochoa@cwi.nl
Thomas Degueule
thomas.degueule@cwi.nl
Jurgen Vinju
jurgen.vinju@cwi.nl
18
11 OSGi best practices
Best practices are not widely
followed
Best practices impact in classpath size
(0-23%) and performance (±13%)
“Modularity without regret.”