Yes, nothing will make you appreciate the relational model more than using a database where the designer broke first normal form. Or where the developer thought the data in the database was theirs and it was only ever going to be used by just their application.
Something else I find confusing is that every developer seems to want every database query for a single object to return a single row with a thousand columns, and anything multi row or with multiple results is way too complicated to handle. This goes double for report writing software.
I really wonder what we're doing with database providers (drivers) that makes people want to develop ORMs when ORMs consistently feel like 10 ton gorillas. If the database world is so disparaging of the "row by excruciating row" processing, why do the drivers exclusively present data that way and no other?
I can't figure out why when I do `t1 left join t2` and it's not 1:1 or 1:0, I get duplicate t1 rows. Why can't either the driver or SQL find a way to represent a single t1 with multiple t2 (this is of course trivial to represent in JSON or any programming language with arrays and dicts).
I don't mind that the tables in relational databases are essentially 2d matrices. I can do the hierarchy with the joins. I just need a way to get the data out in a sane format.
Last time I brought this up ~15 years ago everyone hated on me. But everyone knows this is ultimately the data we want to get out of the system.
I mean, I understand why it does that currently. It's because query results are always relations (tables), and relations are defined as an unordered set of records. By definition you can't use row order meaningfully in a relation, and by extension row relationships. So it duplicates the t1 rows in order to show which records from t1 correlate to the records in t2.
However, the issue is that relations are great inside the RDBMS, but they are not necessarily the best presentation of data outside the RDBMS. MS Access, as maligned as it is, has been able to display subsheets in a collapsible hierarchy [0] for decades at this point.
Really, there should be a way to tell the RDBMS to output data as hierarchies based on the join without having to jump through the, quite frankly, completely ridiculous subquery syntax bullshit that JSON and XML formatting currently requires. Even if it only works in 95 out of 100 cases; even if you have to manually tell the query somehow what the key fields are to agglomerate based on. It should be way easier than having to stitch strings together like you're still using printf() for everything at all times.
Something else I find confusing is that every developer seems to want every database query for a single object to return a single row with a thousand columns, and anything multi row or with multiple results is way too complicated to handle. This goes double for report writing software.
I really wonder what we're doing with database providers (drivers) that makes people want to develop ORMs when ORMs consistently feel like 10 ton gorillas. If the database world is so disparaging of the "row by excruciating row" processing, why do the drivers exclusively present data that way and no other?