Working quietly: nothing to deploy, a lot in the oven (devblog, week 39)

· all posts

After last week’s long post, this one is short, because this week there is not much we can show you yet. The first beta users started to come in through the new doors (welcome!), and we spent the week heads down on three big things: plugins, Git integration, and speed. None of them is ready for you today, so this is a “what we are cooking” post.

Built, but not deployed yet

Before the kitchen tour: a batch of features is finished, tested and merged into our main line. It is waiting for the next release, so it is not in production yet.

  • Deleting a project or a team is no longer final. It is kept for 30 days before it is really gone, and the delete dialog makes you type the project’s slug to confirm, GitHub style. If you delete something by mistake, write to us within those 30 days and we can bring it back.
  • Files can be renamed, moved and deleted from the project page. A deleted file goes to a trash for 30 days instead of disappearing. The part we are happiest with: when you move or rename a file, the references to it follow. Rename a sheet file and the parent schematic points to the new name; move a footprint library and the library table is updated; 3D model paths follow too. If someone has the file open, the dialog tells you who, and their editor shows a “this file is gone” banner instead of editing something that no longer exists.
  • Everything from the last post’s “in the queue” list: follower counts on Explore, the two drift fixes, the “Add / Remove” library wording, and the newsletter opt-in in the app. (The newsletter form on the landing page is already live, the website ships separately.)
  • Another sync audit. This time it was about schematics with several sheets, and the drawing sheet editor. Five ways two people could end up seeing different things, all five reproduced and fixed, each with a regression test.

Plugins, close to release

Last week the plugin system was “in preview, on staging”. It still is, but it moved a lot closer to something you can use.

  • A Plugins page in the app. You browse the plugins we have reviewed and published, install one with a click, and it shows up in the editor’s plugin menu. Updates are offered, never installed silently, and if a new version asks for more permissions than the old one, you have to approve those again.
  • KiCad 10’s Remote Symbols, in the browser. Desktop KiCad 10 has a panel where a parts provider’s website runs inside KiCad: you pick a part there, and KiCad downloads the symbol and places it. In our WebAssembly build that menu item is grayed out, because there is no embedded browser to run it in. So we rebuilt the KiCad side of it on top of the plugin system, keeping the provider protocol the same. A provider written for desktop KiCad works here too with some easy-to-follow migration steps (for now, only providers that don’t need login). The selected part lands in a team library first, so your teammates get it as well.
  • Plugins can ask for manufacturing files. Gerbers, drill files, an IPC-356 netlist and a component list, generated on our servers and zipped. This is the groundwork for ordering-style plugins.
  • Plugins start faster: the logic and the UI now load in parallel, and opening a plugin panel is one round trip instead of several.

Plugins stay switched off in production until we are happy with the security story.

Git: working, now making it smooth

In the last post Git integration was a design document “on paper”. By the end of this week most of it was built on staging, behind a preview switch that only we can see.

What works:

  • A portable export: a zip that opens in desktop KiCad 9 or 10 with the libraries it needs bundled, and that you can upload back with a preview of what changed.
  • Comments survive branches. Threads moved into one document per project, so switching branches never loses a thread, and each branch shows whether a comment’s item is still there.
  • @mentions in comments now send a notification, with an inbox to read them.
  • Connect a project to a Git repository (any Git host, GitHub is a convenience, not a requirement) and check out a branch. A branch opens as a working copy: if a teammate picks the same branch, you edit it together, live, like any other project.
  • Commit the files you choose, with a summary per file that says who changed what, then push, create branches and tags, and pull in changes from the remote.
  • Libraries can have their own repository, independent of the project: pull a team library from Git, push its changes back.

What does not work well yet is the feel. It is a big feature, and the first version of every screen shows it. This week and the next go into the UI and UX pass (a new “new project” screen that can start from a Git URL is part of it).

No in-app conflict resolution and no force pushes. That was the plan, and it stays the plan.

Speed, and the road back upstream

Our goal is to run pcbjam on a mid-range Android phone or tablet. Today a big board is laggy even on a MacBook Pro. So we started digging into the KiCad codebase itself: finding the places where it is slow in the browser, and checking which ones can be made faster and which ones are already as good as they get.

Our usual load test has been our DDR5 tester board. The board we measure on now is bigger than that, and the early results are very promising: panning, zooming, routing, dragging a footprint and DRC all got a lot faster in our experiments. We are not putting numbers on it yet, because it is work in progress and we want to measure it properly first. For a teaser; 5x is a conservative guess.

The plan after that is a rebase of our KiCad fork onto current mainline KiCad, and then opening pull requests upstream with the speedups, so desktop KiCad users get them too. That part is not as simple as it sounds: our fork has diverged in some files, so some changes will have to be rewritten against upstream instead of simply cherry-picked. We would still want to do the work because it is good for the whole community.

Weekly cliché

“Quiet weeks are the ones that make the loud ones possible.” Or at least that is what we tell ourselves.