macro bookcase_alloc_macros

Companion procedural macro crate for bookcase_alloc

2 stable releases

2.0.0 Dec 9, 2022
1.0.0 Dec 5, 2022


Used in bookcase_alloc

MIT/Apache

9KB

📚 Bookcase - An Arena Allocator

I wanted to learn arenas, so I implemented my own. Then I noticed the existing arena crates were not to my liking.

🕐 Project Status

Experimental, do not use unless you are a 🤡.

📖 Glossary

  • Person: thread
  • Notebook: arena
  • Chapter: set of commonly aligned pages
  • Page: fixed sized buffer of raw bytes
  • Utensil: implementation of allocation

Notebooks start out blank. A person can write into a notebook. Depending on if they use pencil or pen, they may or may not be able to erase their notes. A person can pass a notebook to another person. Any person can write in the notebook, but only one at a time. Any number of people can read from a notebook at the same time.

This analogy is not perfect, but it's way better than what 'arena' has to offer! 😁

🎯 Goals

  • Thread safe
  • Safe interface
  • Ergonomic
  • Fast
  • Lean
  • Configurable
  • Building block for data structures
  • No external dependencies
  • Documented

🚫 Concessions

  • Unsafe implementation details
  • Lookup is out of scope (this is not an ECS library)

🚀 Progress

In rough priority order:

  • CI
  • CD
  • No dependencies
  • Well documented
  • Thread-local notebooks (Personal*Notebook: Send)
  • Thread-safe notebooks (Public*Notebook: Send + Sync, very unsafe)
  • Bump allocation (Pen)
  • Deallocation
  • Compiles on stable rust
  • Publish first experimental version
  • Publish first beta version
  • Publish first stable version
  • Heterogeneous notebook (*MultiNotebook: Notebook)
  • Homogeneous notebook (*MonoNotebook<T>: TypedNotebook<T>)
  • All allocations are aligned
  • *MultiNotebook: Allocator (requires nightly)
  • Configurable page base size (SizeStrategy)
  • Configurable page growth rate (GrowthStrategy)
  • Non-dropping exclusive references (alloc*() -> &mut T)
  • Auto-dropping handles (new*() -> Handle<T>)
  • Notebook merging

🌳 Versioning

Philosophy

  1. Backwards compatibility is falling asleep to the sound of ocean waves breaking on the beach.
  2. Assuming backwards compatibility is torturing puppies.

Conclusion: SemVer is 🐍🛢

Consequence

Unfortunately, cargo is tied to SemVer. Fortunately, SemVer is versioned. This means I can create my own version of SemVer! 😈 Call it SemVer Maggie.1.0. Here's how it works:

The Ironclad Rule

Assume all versions are breaking.

The Motto

Strive to keep breaking changes to an absolute minimum.

Release Channels

Each dot separated number represents a release channel:

Stable

This channel is the only one suitable for production use.

Beta

This channel is for collecting well baked ideas that are preparing for stabilization.

Experimental

This channel is the wild west where all bets are off and clown behavior is the norm.

Format: stable.beta.experimental

  1. The version is in the experimental channel when experimental > 0, otherwise...
  2. The version is in the beta channel when beta > 0, otherwise...
  3. The version is in the stable channel when stable > 0, otherwise...
  4. The version is invalid.

Therefore all stable releases are a bump to the "major" version suggesting to the downstream developer they should consider The Ironclad Rule even if they are unfamiliar with SemVer Maggie.1.0.

Examples

  • =1.0.0 - stable v1
  • =1.2.0 - beta v2 based off stable v1
  • =1.2.3 - experimental v3 based off beta v2 based off stable v1
  • =2.0.1 - experimental v1 based off stable v2
  • =0.0.0 - invalid

But I Like bookcase_alloc = "^1"!

Then get accustomed to compiler errors. A version within a release channel will only compile with its respective feature enabled. The default feature is of course stable. Enabling more than one will also fail to compile. This is to prevent accidental use of beta and experimental channels in production user code.

... Maggie?

Maggie is my pet 🐷. She sucks at first, but you will eventually grow to love her.

Exceptions

bookcase_alloc_macros is itself required to enforce the rules of SemVer Maggie.1.0. As such, it will only have a stable release channel.

Contribution

I created a discord channel. As of this writing it has a population of me. Read the Goals, Concessions, and Progress sections for ideas on what to work on, and speak with me about how to make changes I am likely to accept. You can also just ask questions and give me feedback: feature requests, tell me my code is terrible or that I'm being too edgy. All feedback is welcome!

No runtime deps