This repository contains supplementary material for the paper "Execution-Aware Program Reduction for WebAssembly via Record and Replay" (ASE 2025).

• Any OS that supports docker can run this artifact.
• To reproduce the effectiveness and efficiency number of the paper, you might need a hardware matching the performance of the machine used for the original experiment (Intel Core i9-13900k CPU with 5600Mhz 192GB of DRAM)

We provide three ways to install our artifact.

• Run the image: docker run -it doehyunbaek1/rr-reduce:ase_2025

• Install Visual Studio Code and its Dev Containers extension.
• Open the repository in Visual Studio Code: code ..
• Build and open the project with Visual Studio Code Command: Dev Containers: Rebuild and Reopen in Container.

We officially support only X86-64 based Ubuntu 24.04 machines for local installation. Consult .devcontainer directory to find the relevant scripts.

We suppose that you have installed the artifact with one of the three methods described above.

RR-Reduce in the paper is named as wasm-slice in the artifact.

Here is the usage:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ wasm-slice
usage: wasm-slice [-h] oracle_script test_input test_output
wasm-slice: error: the following arguments are required: oracle_script, test_input, test_output

For example, you can run the following command:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ wasm-slice evaluation/oracle/fixed-by-4e3e221.py evaluation/benchmarks/ffmpeg/ffmpeg.wasm reproduced.ffmpeg.sliced.wasm
Input Oracle Path: evaluation/oracle/fixed-by-4e3e221.py
Input Wasm Path: evaluation/benchmarks/ffmpeg/ffmpeg.wasm
Input Wasm Code Size: 5,356,751 bytes
============================ Identifying candidate target functions ============================
Heuristic function indices: [186]
Dynamic function indices (except heuristic): [0, 253, 473, 1966, 5673, 5715, 5721]
All function indices (except heuristic and dynamic): [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...], 5712 remaining
============================ For each target Function ============================
Running ffmpeg - 186: Interesting!
Target function index: 186, identified in heuristic set
============================ Reduced Output ============================
Reduced Output Wasm Path: reproduced.ffmpeg.sliced.wasm
Reduced Output Wasm Code Size: 33,037 bytes (0.62% of input), Target Size: 6,088 bytes (0.11% of input)

You can see that given the oracle script evaluation/oracle/fixed-by-4e3e221.py and input Wasm evaluation/benchmarks/ffmpeg/ffmpeg.wasm with size 5,356,751 bytes, RR-Reduce was able to reduce the code size to 0.62% of input in terms of code size (0.11% of input in terms of target size).

Hybrid-Reduce in the paper is named as wasm-hybrid in the artifact.

Here is the usage:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ wasm-hybrid
usage: wasm-hybrid [-h] oracle_script test_input test_output
wasm-hybrid: error: the following arguments are required: oracle_script, test_input, test_output

For example, you can run the following command:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ wasm-hybrid evaluation/oracle/fixed-by-4e3e221.py evaluation/benchmarks/ffmpeg/ffmpeg.wasm reproduced.ffmpeg.hybrid.wasm
Input Oracle Path: evaluation/oracle/fixed-by-4e3e221.py
Input Wasm Path: evaluation/benchmarks/ffmpeg/ffmpeg.wasm
Input Wasm Code Size: 5,356,751 bytes
============================ Running wasm-slice (RR-Reduce) ============================
============================ Running wasm-reduce ============================
Reduced Output Wasm Path: reproduced.ffmpeg.hybrid.wasm
Reduced Output Wasm Code Size: 45 bytes (0.00% of input)

You can see that given the oracle script evaluation/oracle/fixed-by-4e3e221.py and input Wasm evaluation/benchmarks/ffmpeg/ffmpeg.wasm with size 5,356,751 bytes, RR-Reduce was able to reduce the code size to less than 0.01% of input in terms of code size.

All experiments in the paper can be reproduced by running evaluation/eval_reduce.py script. Please refer to the next section for details. To quickly check its working, run the following command:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ TIMEOUT=3600 python evaluation/eval_reduce.py all ffmpeg
2025-09-25 14:01:34,137 - INFO - Starting experiment... you can also check /workspaces/wasm-r3/evaluation/logs/250925/140134/entry.log for this log
2025-09-25 14:01:34,137 - INFO - MAX_WORKER: 8
2025-09-25 14:01:34,137 - INFO - BINARYEN_CORES: 4
2025-09-25 14:01:34,137 - INFO - TIMEOUT: 3600
2025-09-25 14:01:34,137 - INFO - len(toolset): 4
2025-09-25 14:01:34,137 - INFO - toolset: dict_keys(['wasm-slice', 'wasm-hybrid', 'wasm-reduce', 'wasm-shrink'])
2025-09-25 14:01:34,137 - INFO - len(testset): 1
2025-09-25 14:01:34,137 - INFO - testset: ['ffmpeg']
2025-09-25 14:01:34,139 - INFO - Starting reduction for wasm-slice / ffmpeg, logging to /workspaces/wasm-r3/evaluation/logs/250925/140134/ffmpeg-wasm-slice.log
2025-09-25 14:01:34,140 - INFO - Starting reduction for wasm-hybrid / ffmpeg, logging to /workspaces/wasm-r3/evaluation/logs/250925/140134/ffmpeg-wasm-hybrid.log
2025-09-25 14:01:34,140 - INFO - Starting reduction for wasm-shrink / ffmpeg, logging to /workspaces/wasm-r3/evaluation/logs/250925/140134/ffmpeg-wasm-shrink.log
2025-09-25 14:01:34,140 - INFO - Starting reduction for wasm-reduce / ffmpeg, logging to /workspaces/wasm-r3/evaluation/logs/250925/140134/ffmpeg-wasm-reduce.log
2025-09-25 14:01:49,402 - INFO - Finished reduction for wasm-slice / ffmpeg: Elapsed time: 15.21385669708252 seconds, Code size: 33,037, Target size: 6,088
2025-09-25 14:03:57,114 - INFO - Finished reduction for wasm-hybrid / ffmpeg: Elapsed time: 142.95170974731445 seconds, Code size: 45
2025-09-25 14:15:32,657 - INFO - Finished reduction for wasm-reduce / ffmpeg: Elapsed time: 838.4933569431305 seconds, Code size: 479
2025-09-25 15:01:34,580 - INFO - Finished reduction for wasm-shrink / ffmpeg: Elapsed time: 3600.004099369049 seconds, Code size: 4,039,110
2025-09-25 15:01:34,582 - INFO - Took 3600.444494009018s (1.0001234705580606h)

This runs all four tools (wasm-shrink, wasm-reduce, wasm-slice, wasm-hybrid) on evaluation/benchmarks/ffmpeg/ffmpeg.wasm with a timeout of one hour (3600 seconds) for each tool. For wasm-slice, wasm-hybrid, and wasm-reduce, the tools finished within the timeout and the code size agrees with the expected output. For wasm-shrink, as it timed out after 1 hour, the output is much larger than the expected ouput. For full experiment, please refer to the next section.

NOTE: elapsed time might differ depending on your hardware setup.

We provide one script to reproduce the experiments of the paper and the other script to generate the tables and figures used in the paper.

evaluation/eval_reduce.py is the main script used for running all the experiments in the paper.

Its usage is as follows: python eval_reduce [<tool>|all] [<test>|all]

For <tool>, you can provide one of the four tools listed below:

• wasm-shrink
• wasm-reduce
• wasm-slice (Correspondds to RR-Reduce in the paper)
• wasm-hybrid (Corresponds to Hybrid-Reduce in the paper)

For <test>, you can provide one of the 28 programs listed below:

• wasmedge#3018
• wamr#2789
• wasmedge#3019
• wamr#2862
• wamr#2450
• wasmedge#3076
• mandelbrot
• pathfinding
• pacalc
• wasmedge#3057
• guiicons
• rtexviewer
• rfxgen
• riconpacker
• rguistyler
• rguilayout
• jqkungfu
• bullet
• funky-kart
• sqlgui
• hydro
• figma-startpage
• sandspiel
• parquet
• commanderkeen
• jsc
• boa
• ffmpeg

Output of the metric will be written in evaluation/metrics.json file. Output of the original metric is checked in at evaluation/expected_metrics.json.

Output of the individual reduction will be written inside directory evaluation/benchmarks/<test>/. For example, here is what evaluation/benchmarks/commanderkeen/. would look like:

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ ls -l evaluation/benchmarks/commanderkeen/
total 29436
-rw-r--r--  1 vscode vscode   363027 Sep  4 21:49 commanderkeen.hybrid_stripped.wasm
-rw-r--r--  1 vscode vscode   363027 Sep  4 21:49 commanderkeen.hybrid_temp.wasm
-rw-r--r--  1 vscode vscode      951 Sep  4 23:36 commanderkeen.hybrid_test.wasm
-rw-r--r--  1 vscode vscode      951 Sep  4 23:36 commanderkeen.hybrid.wasm
-rw-r--r--  1 vscode vscode  1690562 Sep  7 05:08 commanderkeen.reduced_test.wasm
-rw-rw-r--  1 vscode vscode  1690577 Sep  7 05:08 commanderkeen.reduced.wasm
-rw-------  1 vscode vscode  1969892 Sep 11 07:30 commanderkeen.shrunken.wasm
-rw-r--r--  1 vscode vscode 14986192 Sep 11 07:30 commanderkeen.shrunken.wat
-rw-r--r--  1 vscode vscode   387482 Sep  4 14:10 commanderkeen.sliced.wasm
-rw-r--r--  1 vscode vscode  4331018 Sep  4 21:32 commanderkeen.stripped.wasm
-rw-r--r--  1 vscode vscode  4331018 Sep  4 21:32 commanderkeen.wasm
drwxr-xr-x  2 vscode vscode     4096 Sep 12 07:39 expected
drwxr-xr-x 16 vscode vscode     4096 Sep  4 14:10 out

• commanderkeen.wasm is the input to the reduction tools.
• commanderkeen.shrunken.wasm is the output of the wasm-shrink tool
• commanderkeen.reduced.wasm is the output of the wasm-reduce tool.
• commanderkeen.sliced.wasm is the output of the wasm-slice (RR-Reduce) tool.
• commanderkeen.hybrid.wasm is the output of the wasm-hybrid (Hybrid-Reduce) tool.
• commanderkeen.hybrid_stripped.wasm, commanderkeen.hybrid_temp.wasm, commanderkeen.hybrid_test.wasm, commanderkeen.reduced_test.wasm, commanderkeen.shrunken.wat, commanderkeen.stripped.wasm, are the intermediate files generated by the tools.
• Directory out contains intermediate files generated by wasm-slice and wasm-hybrid.
• Directory expected contains output of the original run for comparison.

NOTE: the size you get with ls -l is the size of the module, not the code. To get the code size you need to use tools like wasm-tools objdump or consult metrics.json which utilizes such tool.

For reference, we provide how long it took to reproduce the experiment with each tool.

NOTE: your result might differ depending on your hardware setup.

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ python evaluation/eval_reduce.py wasm-shrink all
2025-09-12 07:30:34,200 - INFO - Took 172802.76957297325s (48.0007693258259h)

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ python evaluation/eval_reduce.py wasm-reduce all
2025-09-08 01:53:27,849 - INFO - Took 161088.80288028717s (44.74688968896866h)

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ python evaluation/eval_reduce.py wasm-slice all
2025-09-04 19:09:01,172 - INFO - Took 18766.46604323387s (5.212907234231631h)

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ python evaluation/eval_reduce.py wasm-hybrid all
2025-09-05 21:53:01,421 - INFO - Took 87603.23522043228s (24.334232005675634h)

To reproduce the tables and figures used in the paper, you run evaluation/latexify.py.

vscode ➜ /workspaces/wasm-r3 (ASE_2025) $ python evaluation/latexify.py evaluation/metrics.json

The tables and figures will be written to evaluation/table/ and evaluation/figure/.

The Repository is structured as follows:

• ./rr-reduce: This directory contains Python code used in the implementation.
  • ./rr-reduce/wasm-slice: This file is a entry point for RR-Reduce
  • ./rr-reduce/wasm-hybrid: This file is a entry point for Hybrid-Reduce
  • ./rr-reduce/heuristic_finder.py: This file contains logic for identifying Heuristic Set.
• ./src: This directory contains JavaScript code used in the implementation.
  • ./src/test.ts: This file is a entry point to Wasm-R3.
  • ./src/test-slice-dice.ts: This file calls other scripts that does split, record-replay, and combine.
  • ./src/web.ts: This file implements record phase and reduce phase of Wasm-R3.
• ./crates: This directory contains Rust code used in the implementation.
  • ./crates/replay_gen: This directory contains replay phase of Wasm-R3.
  • ./crates/slice_dice: This directory contains the split program phase of RR-Reduce.
• ./evaluation: This directory contains evaluation set and evaluation scripts.
  • ./evaluation/benchmarks: This directory contains 28 Wasm programs used for the evaluation of RR-Reduce. We also include output of RR-Reduce (suffixed with .sliced.wasm), Hybrid-Reduce (suffixed with .hybrid.wasm), wasm-reduce (suffixed with .reduced.wasm), and wasm-shrink (suffixed with shrunken.wasm).
  • ./evaluation/engines: This directory contains different versioins of three Wasm engines used for the evaluation.
  • ./evaluation/oracle: This directory contains oracle scripts used for the evaluation.
• ./third_party: This directory contains third party dependencies used in the implementation.