o3 estimates 3300 man hours to reach feature parity with CheerpX, wdyt? #1316

OhadRubin · 2025-04-26T16:23:06Z

OhadRubin
Apr 26, 2025

I asked a language model what will it take to reach feature parity with CheerpX and this is what he told me

Gap inventory

Area	What CheerpX offers	What v86 lacks	Typical work items
JIT throughput	5-10 × native-slower WebAssembly JIT (CheerpX 1.0: high performance x86 virtualization in the browser via ...)	30-100 × native; simpler trace cache	redesign code generator, profile-guided re-compilation
ISA coverage	full 32-bit x86 incl. MMX/SSE2 (CheerpX Documentation)	MMX, large parts of SSE/FPU missing (Pixelsuft/v86-wasmless: An old non-wasm version of Copy's Virtual ...)	add decoding, micro-ops, tests
Multiprocess / fork	Linux syscall emulator, works with multi-process programs ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	only single-CPU, partial syscalls
Networking UX	lwIP + built-in Tailscale TCP/UDP stack ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	NE2000 + user-supplied SLiRP
Persistent storage	block filesystem + IndexedDB cache ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	raw image download, no snapshot API
Browser integration	packaged loader, dynamic WASM memory growth	manual HTML wiring	API refactor, npm wrapper, docs
Test & CI matrix	full unit and guest-OS boot tests (implicit)	limited device and API tests (v86/tests/Readme.md at master · copy/v86 - GitHub)	extend harness, headless browsers

Conversion

3 300 h ≈ 20 man-months ≈ two senior engineers for one year.

Uncertainty

±30 %: CheerpX is closed-source; exact internal features unknown.
WebAssembly host limits (2 GB memory, main-thread JIT pause) impose ceilings that even feature parity cannot remove.*

This estimate presumes reuse of v86 architecture and existing community code; rewriting from scratch would at least double the hours.

### Gap inventory

Area	What CheerpX offers	What v86 lacks	Typical work items
JIT throughput	5-10 × native-slower WebAssembly JIT ([CheerpX 1.0: high performance x86 virtualization in the browser via ...](https://www.reddit.com/r/programming/comments/1h6h3bs/cheerpx_10_high_performance_x86_virtualization_in/?utm_source=chatgpt.com))	30-100 × native; simpler trace cache	redesign code generator, profile-guided re-compilation
ISA coverage	full 32-bit x86 incl. MMX/SSE2 ([CheerpX Documentation](https://cheerpx.io/docs/overview?utm_source=chatgpt.com))	MMX, large parts of SSE/FPU missing ([Pixelsuft/v86-wasmless: An old non-wasm version of Copy's Virtual ...](https://github.com/Pixelsuft/v86-wasmless?utm_source=chatgpt.com))	add decoding, micro-ops, tests
Multiprocess / fork	Linux syscall emulator, works with multi-process programs ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	only single-CPU, partial syscalls
Networking UX	lwIP + built-in Tailscale TCP/UDP stack ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	NE2000 + user-supplied SLiRP
Persistent storage	block filesystem + IndexedDB cache ([WebVM	WebAssembly Virtualisation - Leaning Technologies](https://leaningtech.com/webvm/?utm_source=chatgpt.com))	raw image download, no snapshot API
Browser integration	packaged loader, dynamic WASM memory growth	manual HTML wiring	API refactor, npm wrapper, docs
Test & CI matrix	full unit and guest-OS boot tests (implicit)	limited device and API tests ([v86/tests/Readme.md at master · copy/v86 - GitHub](https://github.com/copy/v86/blob/master/tests/Readme.md?utm_source=chatgpt.com))	extend harness, headless browsers

Rough effort model

Work-package	Person-hours	Notes (competent engineer, full-time)
Advanced JIT (baseline → 5-10× native slowdown)	1 200	tier-1 ISA, hot-path caching, invalidation
Complete ISA + FPU/SSE tests	400	includes regression automation
SMP & process support	600	WASM threads, atomic + scheduler work
Userspace TCP/UDP + Tailscale bridge	300	port lwIP, WebRTC tunnel, UI
Persistent FS layers	200	IndexedDB driver, quota, snapshot
Packaging, docs, demo site	200	npm, CDN, examples
QA, fuzzing, cross-browser CI	400	add to GitHub Actions, playwright
Total	≈ 3 300 h

Conversion

3 300 h ≈ 20 man-months ≈ two senior engineers for one year.

Uncertainty

±30 %: CheerpX is closed-source; exact internal features unknown.
WebAssembly host limits (2 GB memory, main-thread JIT pause) impose ceilings that even feature parity cannot remove.*

This estimate presumes reuse of v86 architecture and existing community code; rewriting from scratch would at least double the hours.

copy · 2025-05-23T14:28:06Z

copy
May 23, 2025
Maintainer

Huh

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

o3 estimates 3300 man hours to reach feature parity with CheerpX, wdyt? #1316

Uh oh!

{{title}}

Uh oh!

Replies: 1 comment

Uh oh!

{{title}}

Uh oh!

Select a reply

Uh oh!

o3 estimates 3300 man hours to reach feature parity with CheerpX, wdyt? #1316

Uh oh!

OhadRubin Apr 26, 2025

Conversion

Uncertainty

Rough effort model

Conversion

Uncertainty

Replies: 1 comment

Uh oh!

copy May 23, 2025 Maintainer

OhadRubin
Apr 26, 2025

copy
May 23, 2025
Maintainer