The project seems to have some big problems though. The last stable release was three years ago. 100% reproducible regression bugs for very simple input are not even looked at for years. The build process is so complicated that even the project itself has given up on creating Windows binaries.
The website with the documentation is often down or extremely slow.
The dot graph description language is frustratingly non-composable. Clusters can just barely be used as a crude, limited workaround. A declarative styling mechanism instead of the current imperative state machine approach would remove so much pain.
It's also quite easy to run into crashes.
And still there's nothing better.
I beg to differ, although I'm working for a developer of a (quite costly) commercial alternative. But the state in the open-source or free space has only few good options and many of them are sorely lacking in quality.