* If the form deals with state and validation, then <Button> (as opposed to <button>) can automatically disable itself if the form's input isn't valid.
* It can incorporate useful features like preventing redundant clicks (to prevent duplicate submits) and showing submit progress (eg., replace button label with a spinner while it's pending).
* You can more easily add features that you would have to manually wire with an HTML <button>: For example, let it take a "tooltip" prop that will automatically show a tooltip on hover.
As an aside, I should add that I wouldn't code up a "form with submit and cancel" like this. I would have the form component itself render the submit and cancel buttons, and call it something like SubmittableForm. This way, it can be in control of the UI (eg., validation, button placement, etc.) and submit logic.
As for that htmlFor thing, I don't know why this library does it that way. There's no reason it couldn't correctly wire up a "for" attribute itself on the HTML label element.
Next, you might want to dynamically change the button's text, color, visibility etc. At that point, if you are using pure DOM API, the code will start to look like spaghetti, unless what you're doing is a really simple one-off form page.
Things like button's text, color, visibility is nicely controlled via adding/removing classes. And this <Button /> element can create enough spaghetti of its own, too, with attributes like "size", "type", etc, if overused. And with some attributes it can be hard to distinguish native html attributes from component-specific attributes.
Isn't the whole point of using React that I don't have to manually manage state like this? When text, color, visibility are all changed independently in different places for more than a handful of states, suddenly it isn't so nicely controlled.
I assume you'd learn it because it fits with the majority of your use case; having labels immediately attached to fields. So not only are you enforcing a particular style but your saving keystrokes and mental energy on two tags when you only need one.
The thing with html, you're gonna start fighting it sooner or later. And when this moment comes, i would prefer to find solutions to the wide-known html problems, not to the problems specific to this library, which may kind of create them in the first place.
One of the advantages of React compared to Angular is that you don't have to learn specific angular syntax — you are encouraged to use native javascript as much as you can. Same goes for html — i'd like to use native html as much as I can. It has enough problems of its own, but at least I know those problems, and which I don't, i'm pretty much sure i can google solution for.
But these component-wrappers are not solving html problems, they are creating shortcuts which are definitely shorter, but rarely intuitive and introduce new syntax and may introduce new problems.
Take, for example, the <FormField label="Input name" htmlFor="input-name" /> component. It takes two props which have totally different meaning. htmlFor is react's substitute for the native html "for" attribute, but the label is being created as an html element. So I have to learn this new mixed up syntax. Why? Existing html syntax is nice already and what everyone's used to.
On the other side, such elements as <Spinner /> are nice to have. We don't have any native html spinner elements, so it's a good unifying wrapper for arbitrary spinner html+css magic, which can be easily changed once someone on the project wishes to change how the spinners look.