digitalmars.D.announce - LWDR (Light Weight D Runtime) v0.3.0
- Dylan Graham (96/96) Jun 19 [Github](https://github.com/0dyl/LWDR)
- Dylan Graham (5/9) Jun 19 Once LWDR is stable enough, I want the next version to include
- lili (2/2) Jul 09 Great Work! Why standard D Runtime can not run on MCU? Will you
[Github](https://github.com/0dyl/LWDR) [DUB](https://code.dlang.org/packages/lwdr) [Previous announcement](https://forum.dlang.org/post/giigcnoyxfoxxaevjmxy forum.dlang.org) LWDR (Light Weight D Runtime) is a ground-up implementation of a D runtime targeting the ARM Cortex-M microcontrollers and other barebones environments. It works by providing a series of basic API hooks (as defined in [rtoslink.d](https://github.com/0dyl/LWDR/blob/master/source/rtoslink.d) that you must implement and/or point to your RTOS implementation. This is V0.3.0 of LWDR. Since V0.2.3, the following has been worked on: 1. Thread Local Storage support 2. Primitive memory tracking for Phobos allocations that would normally rely on a GC 3. Transition to an opt-in system 4. Replacement of `delete` with `LWDR.free(..)` due to deprecation 5. Source code documentation improvements 6. `RefCount!T` and `Unique!T` LWDR-specific implementations **Thread Local Storage** This feature is rather abstract, and it is an opt-in with version `LWDR_TLS`. You must provide support in your linker script for `tdata` and `tbss` sections. It works by utilising the underlying RTOS's TLS implementation ([example](https://www.freertos.org/pvTaskGetThreadLocalStoragePointer.html)). When `LWDR.registerCurrentThread()` is called a block of D memory is allocated containing the TLS variables for the current thread, and the pointer to the block is stored in the thread's TCB (Thread Control Block). When a TLS variable (ie, a static variable) is accessed, `__aeabi_read_tp` is called, yielding the pointer. **Memory Tracking** This is very primitive. It's *only* meant to assist with stopping GC-reliant stdlib allocations from leaking. It pretty much behaves as defined [here](https://forum.dlang.org/post/gfyhdqecjxszrgutlhmi forum.dlang.org). **Opt In** To be able to keep the size of `TypeInfo` vtables down and such, LWDR has adopted an opt-in system, which relies on [D's version feature](https://dlang.org/spec/version.html). The current opt-ins are: 1. `LWDR_TLS` - Enables TLS support 2. `LWDR_DynamicArray` - Enables dynamic arrays 3. `LWDR_TrackMem` - Enables the mess above. **Replacement of `delete`** `delete` has been [deprecated](https://dlang.org/deprecate.html#delete). `LWDR.free` has been implemented in its place to prevent compiler warnings. **Source Code Documentation** Runtimes are hairy and scary - so I'm beginning to put more effort into documenting how things work. So far, it's only [ddoc](https://dlang.org/spec/ddoc.html) comments. **`RefCount!T` and `Unique!T`** To alleviate the lack of GC, I have implemented an LWDR-specific solution inspired by [automem](https://code.dlang.org/packages/automem). **What works?** 1. Class allocations and deallocations (via `new` and `LWDR.free`) 2. Struct heap allocations and deallocations (via `new` and `LWDR.free`) 3. Invariants 4. Asserts 5. Contract programming 6. Basic RTTI (via `TypeInfo` stubs) 7. Interfaces 8. Static Arrays 9. Virtual functions and overrides 10. Abstract classes 11. Static classes 12. Allocation and deallocation of dynamic arrays (opt in by version `LWDR_DynamicArray`) 13. Concatenate an item to a dynamic array (opt in by version `LWDR_DynamicArray`) 14. Concatenate two dynamic arrays together (opt in by version `LWDR_DynamicArray`) 15. Dynamic array resizing (opt in by version `LWDR_DynamicArray`) 16. Thread local storage (opt in by version `LWDR_TLS`) **What doesn't work?** 1. Exceptions and Throwables (experimental implementation was removed) 2. Module constructors and destructors 3. Static constructors and destructors 4. Shared static constructors and destructors 5. Module info 6. There is no GC implementation (primitive memory tracking is now available with `LWDR_TrackMem`, `RefCount!T` and `Unique!T` are now available) 7. Delegates/closures 8. Associative arrays 9. Object monitors 10. `shared`/`synchronised` 11. Object hashing 12. Other things I can't remember off the top of my head. It's still a beta - so expect bugs and warts. Some bits have been thoroughly tested, others not so much. Because the runtime has ballooned so quickly, I want to pause on development for a little bit so that I can begin using LWDR in a proper project and find and squash bugs. The project is pretty much the successor to my [Driving with D article](https://dlang.org/blog/2021/06/01/driving-with-d/) (it's an automotive project). I'm also thinking of applying with LWDR to the Autumn of Code thingo.
On Saturday, 19 June 2021 at 13:31:11 UTC, Dylan Graham wrote:[Github](https://github.com/0dyl/LWDR) [DUB](https://code.dlang.org/packages/lwdr) [Previous announcement](https://forum.dlang.org/post/giigcnoyxfoxxaevjmxy forum.dlang.org)Once LWDR is stable enough, I want the next version to include module info, static ctor/dtor support and to look into Object monitor support and other multithreading tools (Mutexes, Conditions, etc).
lili <akozhao tencent.com> writes:
Great Work! Why standard D Runtime can not run on MCU? Will you plan that LWDR support other arch like RISC-v ?
rikki cattermole <rikki cattermole.co.nz> writes:
On 10/07/2021 2:30 AM, lili wrote:Why standard D Runtime can not run on MCU?It requires things like threads. Basically it assumes there is a kernel. Anyway, you don't want the regular runtime on a MCU, a MCU is too small in memory to really hold it all and your own logic and any libraries you need for the hardware its connected to.
On Friday, 9 July 2021 at 14:30:07 UTC, lili wrote:Great Work!Thanks!Why standard D Runtime can not run on MCU?The standard D Runtime is reliant on a fully-fledged OS, which don't fit onto small embedded devices and [they're incompatible with them](https://electronics.stackexchange.com/a/19243) For example, Linux requires a memory management unit (MMU), yet the Cortex-M series lack that feature. The runtime itself is quite heavy due to the enormity of the features it supports. For example, the functions in DRuntime's [`lifetime.d`](https://github.com/dlang/druntime/blob/15b49cc725df280a72508f729ec4495ffbac66c3/sr /rt/lifetime.d#L86) are dependent on GC code, which adds memory overhead to store those instructions and stack space. In LWDR, [`lifetime`](https://github.com/hmmdyl/LWDR/tree/master/source/lifetime) functions just point to your own implementation, yielding a leaner memory footprint. It's a very pedantic example, but it adds up after a while on a resource-constrained platform.Will you plan that LWDR support other arch like RISC-v ?Yes :) Asides from some ARM EABI specific stuff in [`rt/sections.d`](https://github.com/hmmdyl/LWDR/blob/d55a39d028942f0eb1582d473317f89030995703/source rt/sections.d#L38), I don't think it'd difficult getting LWDR running on another MCU platform (as of v0.3.0).