GitHub Discussions

Long-form, searchable conversation that future users will find via Google — questions with reusable answers, design proposals before a PR, and write-ups of work built with TensorMesh.

Use Discussions when you want the conversation to stick around. For a fast back-and-forth, hop into Discord instead; for a confirmed bug or a concrete feature request, open a GitHub Issues.

Go to Discussions

Open Discussions on GitHub  →   

You’ll need a GitHub account to post or comment. Reading is open to everyone.

Category guide

There are five categories, each tuned for a different kind of post. Pick the one that matches what you’re writing; if you’re unsure, General is always safe — a maintainer will move it if needed.

Category	Format	What goes here
Announcements	Announcement (read-only)	Release notes, API deprecations, scheduled breaking changes, community events. Maintainer-posted. Subscribe to this category if you depend on the library in production or in a paper.
Q&A	Q&A (with Mark as answer)	Usage questions, install / environment problems, “how do I do X with TensorMesh”. The OP — or a maintainer — marks one reply as the accepted answer, which makes the thread useful to the next person who hits the same problem.
Ideas & RFCs	Open discussion	Proposals for new features, API design, breaking changes — before you write the PR. Floating an idea here first usually saves a round of “we’d actually want this shaped differently” review later. Link the eventual issue / PR in a follow-up comment.
Show & Tell	Open discussion	Papers that used TensorMesh, simulation animations, blog posts, teaching materials, comparison plots against other FEM stacks. Self-promotion is welcome here — that’s the whole point of the category.
General	Open discussion	Anything FEM-, PyTorch-, or PDE-adjacent that doesn’t fit above. Catch-all.

Writing a good Q&A post

The same checklist from the Discord help-channel guide applies, and matters even more here because future readers will arrive at your thread via search — they only have what you wrote. See Asking a good help question for the full list. The short version:

• What you ran — a minimal, copy-pasteable snippet.

• What you saw — the full traceback, in a code block.

• What you expected — shape, value, or behavior.

• Versions — output of python -c "import torch, torch_sla, tensormesh; print(torch.__version__, torch_sla.__version__, tensormesh.__version__)", plus OS and CUDA version if relevant.

A useful habit: when the issue is resolved, edit the OP to add a one-line “Resolution:” summary at the top, so future readers don’t have to scroll through the back-and-forth. Then mark the answering reply.

Discord vs Discussions vs Issues

The three channels solve different problems. The decision is usually simple once you know what you’re trying to do:

You want to …	Use
chat in real time, get unstuck in a few minutes	Discord (#help / #troubleshooting)
ask a question whose answer should be searchable later	Discussions → Q&A
propose a new feature or API change for discussion	Discussions → Ideas & RFCs
share a paper, demo, or write-up	Discussions → Show & Tell
report a confirmed bug with a repro	GitHub Issues
request a specific, actionable feature	GitHub Issues (after a quick Discussions sanity check on bigger features)
announce / be notified of releases	Discussions → Announcements (watch this category)

When in doubt: post to Discussions, link from Discord. Chat that sounded one-off often turns out to be useful to others later, and moving a Discord exchange into a Discussion thread is a maintainer- friendly thing to do.

Labels

Maintainers may tag Q&A discussions with labels like installation, cuda, meshing, assembly, solver, autograd, or docs to make search filtering possible. You don’t need to label your own post — pick the right category, and the labels follow.

Etiquette

The same norms that apply on Discord apply here: be kind, keep disagreements technical, no spam. A couple of Discussions-specific courtesies on top:

• Search before posting. Many questions have already been asked. GitHub’s in-page search across Discussions is decent; a quick site:github.com/camlab-ethz/TensorMesh/discussions <your terms> on Google is even better.

• One topic per thread. If you find yourself describing two unrelated problems in the same post, split it. Threaded replies on Q&A only work well when the question is single-shot.

• Don’t open a duplicate. If your problem matches an existing thread, comment there instead of opening a new one — even if the existing thread is closed. A maintainer will reopen it or redirect.