Hyperlaunch (Minecraft launcher) - devlog 7

last one!

I’ve fixed a lot of launcher bugs and prepared a minimal GUI for trying the launcher out.

Windows and Linux builds are functional and tested. I can’t confirm Mac builds work without bugs, but the launcher core itself was tested on a Mac Neo around a week ago.

I hope anyone reading enjoys Hyperlaunch, even if the user interface is a bit clunky at the moment.

Hyperlaunch (Minecraft launcher) - devlog 6

i lost my sticky streak

• NeoForge support is done for 26.1 and later.
• Added an instance command for rudimentary management of game instances, you can see a demo of the interface below.
• Modrinth support is good enough for a search -> download flow, as also shown below. Still improving over time.

Hyperlaunch - devlog 5

Modrinth support

Modrinth support is progressing nicely, and Hyperlaunch is now (technically) capable of doing more complex searches than Modrinth App allows.
This means Hyperlaunch is better at using Modrinth than Modrinth itself.

NeoForge support

I’ve started work on supporting the NeoForge mod loader inside Hyperlaunch. This is harder than Fabric for a few reasons, but the main issue is that it requires directly patching Minecraft’s code.
The NeoForge installer handles this all, but I want to bring the patching system inside Hyperlaunch.
This is progressing nicely too.
I’ve had to add a dependency on 7-Zip’s LZMA SDK as part of this.

Issues

• spent like 3 hours diagnosing why decompressing the NeoForge patches was broken only for it to be an accidental cast from an int64 to an int32.
• writing JSON classes so Hyperlaunch can read Modrinth API responses is taking too long

Hyperlaunch - devlog 4.

It’s been a month since my last devlog because I’ve just been unmotivated to continue working. I do really want to finish this though, so I’ll see it through.

This update is mostly about bugfixes. There are no big new features done.

Modrinth support

I spent a good 3-5 hours building out a proper, full featured Modrinth search client, just for the API docs to be wrong, meaning I massively over-engineered a simple system. Really annoying.

Bugfixes

All the optimisations and big changes I made over the past few devlogs meant I had a lot of broken code on my hands. Won’t list everything here, but:

• Nearly all code bound to the UI was outdated/wrong
• Microsoft auth broke
• The code that handles downloading the client would not extract some necessary files
• Listing versions to download broke

Small new features

• Hyperlaunch can now generate Java launch commands, handling downloads/logins/etc. automatically.
• That’s.. it.

No video for this devlog, have some assorted screenshots from development on this devlog’s work.

Hyperlaunch - devlog 3

Hyperlaunch now has the fastest download time of any Minecraft launcher I have tested!
This is due to a number of optimisations done, mostly informed from app profiling (you can see a screenshot below.)
Network activity is also more steady, and should start less connections overall.

Hyperlaunch was slower because parts of the cache system & download system were reading and writing to a lot of files at once.
I used the PerfView profiler to figure out how I could improve the download time after testing Modrinth App and Pandora Launcher and scoring twice as worse against both.
I’ve attached a screenshot of PerfView below, but it’s very ugly to look at.

Changes:

• Many syscall optimisations in the download handler & caching system
• Download workers no longer wait for a file to be written to disk before starting another download
• Started on Modrinth API client
• Some bugfixes in client manifest parsing

Hyperlaunch - devlog 2

Hyperlaunch is a Minecraft launcher that tries to be as efficient as possible without compromise.

Progress

• Caching: I finally made a proper global caching system! Its purpose is minimising the network traffic, both for launch speeds and general efficiency. It can get big at times, so I also added compression, using the Brotli algorithm for speed.
  The cache sits in front of every network request. The cache will try to minimise the cost of the network request where possible. Often, it will just find a cached version instead of using the network.

The cache works by URL. It splits the URL into domain name and resource. Take Hyperlaunch’s Stardance project page as an example; if it were to be
cached, the directory would be stardance.hackclub.com, and the
resource would be projects/28654. The resource name is hashed with
XxHash128, a hashing algorithm that’s not used for storing secrets, then
the file at the URL is cached at .hyperlaunch/cache/<directory>/<resource hash>. If the file came with caching instructions (either ETag or Last-Modified HTTP headers), then those are stored alongside the cached file as separate files (see the image attached.)

• Parallelism: The launcher now handles each task at the same time, instead of in order. This is probably the biggest speedup so far.
• Instances: The launcher can now handle multiple installs as instances! This means multiple setups can coexist, and users can pick between them. Think MultiMC or PrismLauncher.
• CLI: Reworked a lot of the CLI as part of these changes.

Difficulties

• Maybe I’m just new to C#, but I found a lot of the HTTP stuff confusing. One example is using the Last-Modified header. If you want to use this header, you need to access the response.Content.Headers variable. For all other headers, you access response.Headers.
• Brotli compression also seemed unusually difficult. It only works on Stream objects, and the syntax was strange which resulted in a few unintentional logic bugs that I just didn’t know how to spot.

Video and some photos of the cache in action below.

The cache in action

Hyperlaunch

devlog 1

Hyperlaunch is a Minecraft launcher that tries to be as efficient as possible without compromising in any area.
I started this project earlier this year, but I want to finish it under Stardance.

Progress

• Started on instances system for managing multiple installs. Like PrismLauncher or MultiMC.
• Added to, cleaned up, and fixed multiple issues in the client manifest parsing system. The client manifest is what tells the launcher how to do its job, where to download everything, and how to start the game. Mojang sometimes changes the format. In the 4-month gap since last working on this, they changed how Java arguments are handled, and added a way for launch ingredients to only be considered on specific OS version ranges (for example, only allow a certain library on Windows versions newer than 22H2). Those were both really annoying to fix, mostly because the parsing code was messy. (Used AI to split classes across files.)
• Added the ability to automatically download and launch Fabric modded versions. (see a demo below) (i really hope the video works, a cdn link should be below otherwise)

cdn video link