Airspeed Velocity Benchmarks for Dulwich

This directory contains ASV (Airspeed Velocity) benchmarks for Dulwich core operations.

Installation

First, install airspeed velocity:

pip install asv

Running Benchmarks

Quick benchmark of current code:

asv run

Run specific benchmarks:

# Run only large history benchmarks
asv run -b LargeHistoryBenchmarks

# Run with specific parameters
asv run -b "LargeHistoryBenchmarks.time_walk_full_history"

Continuous benchmarking across commits:

# Benchmark last 10 commits
asv run HEAD~10..HEAD

# Benchmark specific commit range
asv run v1.0.0..master

Compare results:

# Compare current commit with master
asv continuous master HEAD

# Compare two commits
asv compare v1.0.0 HEAD

Generate HTML reports:

asv publish
asv preview  # Opens in browser

Benchmark Categories

1. Large History Operations (LargeHistoryBenchmarks)

• Walking full commit history
• Limited history walks
• Path-filtered logs
• Rev-list operations
• Merge base calculations
• Log with patches (porcelain.log - equivalent to git log -p)
• Show commits (porcelain.show - equivalent to git show)
• Diff tree (porcelain.diff_tree - diff between commits)

Parameters:

• num_commits: 100, 1000, 5000
• files_per_commit: 10, 50, 100

2. Disk Object Store (DiskObjectStoreBenchmarks)

• Reading objects from disk (loose, packed, or mixed storage)
• Object existence checks
• Iterating all objects

Parameters:

• num_objects: 1000, 10000, 50000
• storage_type: 'loose', 'packed', 'mixed'

3. Pack File Disk Operations (PackFileDiskBenchmarks)

• Random access to packed objects
• Sequential pack reading
• Pack index loading
• Pack verification

Parameters:

• num_objects: 1000, 10000, 50000
• with_deltas: False, True

4. Client/Server Operations (ClientServerBenchmarks)

• Local protocol clone
• Local protocol fetch
• Local protocol push

Parameters:

• num_commits: 100, 1000
• operation: 'fetch', 'push', 'clone'

5. HTTP Protocol Operations (HTTPProtocolBenchmarks)

• HTTP smart protocol clone
• HTTP fetch-pack operations
• With/without compression

Parameters:

• num_commits: 100, 1000
• use_compression: True, False

6. Pack Negotiation (PackNegotiationBenchmarks)

• Negotiation with varying common history
• Pack generation simulation

Parameters:

• num_commits: 100, 1000, 5000
• common_history_percent: 0, 50, 90

7. Large File Handling (LargeFileBenchmarks)

• Adding large files
• Diffing large file changes

Parameters:

• file_size_mb: 1, 10, 50, 100

8. Branch Operations (BranchOperationBenchmarks)

• Listing branches
• Switching branches
• Merging branches

Parameters:

• num_branches: 10, 100, 1000

9. Status Operations (StatusBenchmarks)

• Repository status with many files
• Index diffing

Parameters:

• num_files: 100, 1000, 5000
• percent_modified: 10, 50, 100

10. Garbage Collection (GarbageCollectionBenchmarks)

• Counting objects with porcelain.count_objects
• Full GC with porcelain.gc
• Repacking with porcelain.repack
• Pruning unreachable objects with porcelain.prune
• Filesystem check with porcelain.fsck

Parameters:

• num_loose_objects: 100, 1000, 5000
• unreachable_percent: 0, 10, 50

11. Repacking Operations (RepackingBenchmarks)

• Full repository repack
• Incremental repacking
• Reference packing
• Pack file optimization/verification

Parameters:

• num_packs: 10, 50, 100
• objects_per_pack: 100, 1000, 5000
• with_deltas: False, True

12. Symbolic Reference Operations (SymrefBenchmarks)

• Reading symbolic references
• Following symref chains
• Setting new symrefs
• Updating symref targets
• Listing all refs including symrefs
• Reading remote HEAD symrefs
• Resolving remote symrefs

Parameters:

• num_refs: 10, 100, 1000
• symref_depth: 1, 2, 4

Benchmark Development

To add new benchmarks:

1. Add a new class inheriting from BenchmarkBase
2. Implement setup() method for test data preparation
3. Add benchmark methods prefixed with time_ for timing benchmarks
4. Use params and param_names for parameterized benchmarks

Example:

class MyBenchmarks(BenchmarkBase):
    params = [10, 100, 1000]
    param_names = ['size']
    
    def setup(self, size):
        super().setup()
        # Setup code
        
    def time_my_operation(self, size):
        # Code to benchmark

Integration with CI

The ASV benchmarks can be integrated with GitHub Actions for continuous performance monitoring.

Tips

1. Use --quick for faster test runs during development
2. Set ASV_USE_CONDA=no to use virtualenv instead of conda
3. Use --show-stderr to debug benchmark failures
4. Results are stored in .asv/results/ as JSON files