A site without a running generator
Yesterday we had files and an idea that Git repository could work as article database. Today I wanted to see how these files become a site. I already have that working on my blog, with a local database and generator applying data on templates. Could do something similar here, only change shape of some parts.
What if generator is a CI/CD script? Something changes in data repository, or templates, and that starts work which produces HTML in output repository. From there it could go to web. No generator waiting somewhere for visitors. They get finished pages.
That question brought repositories out of every corner. Data, layout, generator, output... perhaps another for configurations. There is reason behind that. Next to blog I could have portfolio, same sort of structure but different content, and I may want that it looks differently. Why should each need its own generator?
Still, we started with data, builder which also keeps templates, and output. Separate layout repository can wait. We know where boundary is, do not have to build fence there today.
Then I had to stop us adding too much. One layout, two templates, article and index. Policy links in footer can be hardcoded. Read data structure and show it. No rules about what deserves to be there, no publication dates rearranging things, no extra logic which makes it harder to see basic result.
That worked. Directories became indexes, articles became pages, and I could follow hierarchy down to the song article from this morning. Even a directory with its own text and children has somewhere sensible to put both. Nothing particularly magical on screen... which was quite good result.
After local build came GitHub. Four repositories now, counting Article Studio itself as project home. That one is like solution shell, keeping knowledge and work records around subrepositories it does not track. I liked that shape.
For automation we needed dedicated credentials. I asked what options actually mean before choosing. No need to offer me three buttons while I do not know enough to press one. We went with separate tokens for reading sources, writing output and sending notification, then set them up one at a time.
Manual build passed. Change in data started notification and generation. Change in builder started generation too. Output records which exact versions went in. When nothing changed, it did not invent another commit just to show it was busy.
We have private generated staging output now. No new website published, no drafts declared ready because some machine managed to render them. That part still waits for us.
Enough for one day. The generator already knows what to do when we return with another change.