Can anyone tell how Quill compares to Draft and ProseMirror?
There was a separate comment about Draft, so I answered that one there and will just comment on ProseMirror here.
ProseMirror's main is insight is the need to maintain a document model in parallel with the actual DOM. While most modern editors also do this, ProseMirror also allows their document model to be extended and customized, which only Quill and possibly Draft[1] is also capable of. I think this is the right idea and foundation to build on top of (Quill also made this choice).
I don't think it's fair to judge implementation or specifics beyond its ideas just yet since it is still in the development stages and things can change and improve, as stated on its README:
"NOTE: This project is in BETA stage. It isn't thoroughly tested, and the API might still change across 0.x releases. You are welcome to use it, but don't expect it to be very stable yet."
[1] "Possibly" because Draft uses inline styles and only two levels of content (block and inline) so it's far more limiting.
For example, I should be able to embed a "video object", which represents some content that isn't directly editable, and which may exist somewhere else, and is perhaps internally represented using opaque ID that references some backing store. I want to be able to render this as a nice little video renderer that has a little edit button or a context-menu, and which can be edited to change its size, select a time subrange, subtitles and so on. The document editor should treat it as an opaque object that can render itself and marshal/unmarshal itself somehow, and which can be copied, pasted and moved around. Ideally it should also be possible to offer UI interactions, such as mouse clicks (for resizing).
Does Quill handle this better?
ProseMirror, on the other hand, explicitly defines a document model as part of its API, arranged as a tree (DAG). This implies you can do things like represent LaTeX natively without having to flatten it, thus preserving the original structure. For example, an entire \section or \chapter could be a single node. (I don't know if this works in practice.) In theory, its design doesn't require a hard distinction between "text" and "embeds" because everything is both.
The benefit is that don't need a complicated, potentially lossy translation between the flat and non-flat versions. For example, if you're doing LaTeX, and there's a \title and \author in the document, you could keep this around as non-visible nodes that are preserved in the model and get serialized back when you export the edited document. (To be fair, I don't know if Quill supports invisible embeds, perhaps it does?)
Deltas are a simple, iterable output format for Quill. In practice it is much nicer to deal with this in many use cases than the Parchment tree.