There Is No Such Unit as an Install — Edit It and the Next Connection Already Has It

The last piece said a device hands over its own screen. This one is about how that screen arrives.

The quietest difference in this structure is here.

The unit of distribution in the world of apps

In installed software the unit of distribution is a binary.

Change one word on a screen and the procedure is unchanged. Edit code, build, sign, upload, get reviewed, and the user's device downloads. Which is why "just change that one line" becomes a job measured in days.

This is nobody's laziness. It is the unit. What ships is a thing installed inside the device, so there is no way to change it except to remake the whole thing and send the whole thing again.

Here it is handed over when you connect

In this structure the unit of distribution is a document handed over when you connect.

The device holds its screen as a description, and the client receives it on attach. Fix the description and the next connection already has that screen. No build in between, no signing, no download.

This magazine has a sample about an unattended store. The app is a folder, with no build output. Edit the screen, open it again, and that is the screen — the verification keeps two frames, before and after, side by side. There is no build between them.

So there is no owner of a store either

This is where it joins "the garden and the ground".

If there is an install, a seat appears to manage that install. Someone decides what may be installed, someone reviews, someone sets the share. That seat is what makes the fence possible.

With no install, that seat does not appear. The party that made the device decides its own screen, and that screen goes only to whoever attached to that device. Structurally there is nowhere in between to sit.

Not a question of convenience but a question of whom you go through.

What editing actually looks like

Showing raw source being edited would not be honest here. That a description can be hand-edited and that hand-editing is the practical way to author are different claims. In practice it is built, attached and checked in a tool.

So what this piece shows is the result. It was changed, it was opened again, it is different. The point is what was not in between.

One incident this sample caught

In the same sample, the log and the picture disagreed once. The log said three items were low on stock; the capture said "nothing is low".

The cause was that moving to a new screen starts that screen at its own initial state. Data the previous screen held does not follow. The log was recording the server's fact and the screen was drawing its own, so neither was lying.

After fixing it, a rule went into the verification: check that the log and the capture are from the same moment. The value has to be on the line after the screen change for the run to pass.

"The screen follows the server" can come apart like that, and if you only look for it with your eyes, you miss it.