Exactly like what was the case for Athens, the core features don't feel very refined (glitches in block editing and not good at all search) and the focus is on building arguably unnecessary features.
On a cursory look it seems like Obsidian puts more effort into having a solid core, which I appreciate.
But Logseq is actually FOSS and fits my needs better with its paradigms.
Now, they approach the problem differently, to be sure, but so do vi and emacs and visual studio code. People choose between them for software development tasks though each has a wildly different approach such that you might use more than one.
Speaking of which, I wonder whatever happened to that project.
One nice thing about Roam is that because it doesn't use files, it has true transclusion, both at the level of documents and at the level of individual symbols (tags & links). Because of this, renaming a tag globally for example is trivially easy. (I don't think global tag renaming is possible in Obsidian natively (without plugin) yet.)
It's not tags, but this update brought global renaming of properties out-of-the-box. Properties are the data-fields in frontmatter, the yaml-part which acts as a header in Obsidian-files. Being yaml, makes this rather easy and safe to implement. Tags on the side can occur in freetext, which makes it easy to have false-tags found. Putting some automatism on this is a bit risky I guess.
I agree, especially if you copy paste tweets containing hashtags into your files you could end up with accident collisions.
On the other hand, I refactor my tags a lot so I really needed the feature. I ended up using a simple VSCode language server to piggyback on "rename symbols". It's not ideal but it reduced a lot of friction.