Potential TODOs

[archibate]

Workflows

• Automatic release ([Workflow] Automatic version releases #1674, lead: @rexwangcc)
• Enforce code header (Copyright declaration format issue #1024, TODO: automatically check copyright info existence)
• Codecov or something else to cover C++17 (lead: @archibate)
• Set up pylint ([cli] Setup pylint: run 'ti lint' to obtain linting results #1846, lead: @archibate)
• Set up Cpplint

Python frontend

• Move frontend to Python ([Refactor] [IR] Move frontend to Python and expose CHI to Python #925)
• Blender intergration ([Blender] Blender intergration #1773, Taichi-Blend, lead: @archibate)
• Move to multi-pass Python AST transformer (lead: @archibate)
• Possible Python 2.7 support ([Py27] Possible Python 2.7 support? #1842, lead: @archibate)

IR system

• Expose CHI C++ API
• Remove get_current_program ([refactor] [ir] IR system refactorings #988)
• IR standardization ([Refactor] IR standardization #689)
• Remove C++ exceptions
• Refactor gather_statements
• Real function support ([IR] Function definition/function call support in Taichi IR #602, lead: @archibate)

Sparse computation

• hash SNode
• bitmasked garbage collection
• ti.append support multiple child ([Sparse] Make ti.append() support dynamic snode with **multiple child** #1282, lead: @archibate)
• Non-POT internal nodes are not allowed

Backends

• Create a fresh new Context for args+earg+rets per-launch ([Refactor] Create a fresh new Context for args+earg+rets per-launch #1739, lead: @k-ye)
• JS/WASM backend (Taichi.js, lead: @archibate)
• AMDGPU backend (lead: @masahi)
• Restore vectorization on LLVM CPUs (lead: @yuanming-hu)
• Improve float atomic performance on LLVM CPUs ([CC] [autodiff] Support AdStack on C backend #1752)
• Multi-GPU or distributed computation support

(Feel free to add/lead items as you wish.)

[yuanming-hu]

First of all, thank you for putting together this list.

However, I don't think items here can be realistically implemented in a minor version. Prioritization and making realistic plans are vitally important.

Also note that we are still working on v0.7 (#677). Let's keep things simple and stick to one roadmap at a time.

Long-term planning (including creating roadmaps) is my job (instead of someone else's). Having a single, consistent person making crucial decisions is beneficial to the health of the Taichi project.

Thanks again, but I have to close this issue to avoid confusing other developers. After v0.7 is released, I'll open a new roadmap for v0.8.