Making PRs against a feature branch still has the same problem.

Say you're working on a major new feature, like adding a new log in flow. It's a good idea to split it into many small changes: create initial log in form, implement user sessions, add SSO support, add "remember me" functionality, etc.

Those changes will likely depend on each other, for example adding the "remember me" checkbox requires the form to actually be built, and you probably don't want to wait for the reviewers to review one change before continuing your work. This means you naturally end up with PRs that depend on each other in a chain.

Stacked PRs (or stacked diffs, stacked MRs, whatever your company calls it) means that:

1. The code review tool lets you specify dependencies between the PRs, for example the "remember me" PR depends on the initial login form implementation
2. It shows the dependencies visually in the UI, like a chain or tree
3. Changes can't be landed until the PRs they depend on have been reviewed
4. There's a way to land an entire stack
5. When you review a PR later in the stack, it doesn't show any of the changes that were made earlier in the stack. Each PR focuses just on the changes in that part.
6. For each PR, the build steps and linters run for all the changes in the stack up until that point. It helps detect if incompatible changes are made earlier in the stack.

Making all your commits directly to a branch then creating a PR for the whole branch is similar, but reviews are a nightmare since it's only a single review for the entire branch, which can be thousands of lines of code.

At my workplace, we use feature flags (essentially if statements that can be toggled on or off) rather than feature branches. Everyone works directly from the main branch. We use continuous deployment which means landed code is quickly pushed to production. New features are hidden behind a feature flag until they're ready to roll out.