Databases on the other hand are built for this. After years of dissatisfaction with plain text accounting and many hours spent trying to improve it, I now use SQLite and it has been an enormous improvement.
For me, I found that the SQLite models of GnuCash aren't straightforward to query. That's why Beancount created its own query language. Martin Blais has a good discussion of why a traditional database doesn't quite fit for many accounting purposes https://beancount.github.io/docs/beancount_query_language.ht...
Plaintext is nice for git but I only feel that when fixup-ing a single or small number of transactions. It does feel nice to be have all the details of a transaction in one place in a visually useful way. For one-off hacking and such it definitely feels easier to write O(n^4) python looping over trying to describe things with SQL and working at a scale where it doesn’t matter.
Plaintext as a UI into a SQL store seems an interesting project. I would love a git integration for committing changes after diff review and being able to stage individual txns or parts. Many years ago I was frustrated with ledger’s more loosey goosey syntax and trying these things and eventually gave up whatever the idea was at the time. I like the idea of a constant bidrectional sql<->plaintext that provides a requirement for reproducible parsing and serializing
> parsing it in a structured way is a nightmare
Ah, well that’s the job of the PTA app - converting “just text” to something very structured and validated. Which can then be moved into SQLite, if one likes.
> and scripting edits to plain text is a mess.
I suppose it depends. There are a lot of very powerful and quick tools and techniques for automated or assisted text munging.
That gives me the benefits of Fava and all the other PTA tools as well.