as i said, i don't have a problem with people using OIDC if it's a good fit for them. i see that it can be used without falling afoul of these issues, so i'm not here to tell anyone what they should or shoudn't use. if you want to know what *i* prefer, it's home-rolled auth. i hear you about the high-value secret, but i think the "don't store the password in plaintext" failure case of home-rolled auth is much easier to avoid than "don't fuck up your handling of these 4 magic numbers at any point in our 7-step process" in OIDC. and that's only considering the failure case on your side, you now also have to worry about failures and malicious behaviour from the identity provider.
in my experience of using OIDC in the work project i discussed in the post, compared to the home-rolled auth we had before (granted some of these problems are due to the particularly poor choice of identity provider from higher ups):
- now our service's availability is not only dependent on our own uptime, but a third party that has worse uptime than us, which we can't do anything about
- when there is an outage, people can't switch to other tasks because all the services are unavailable, not just ours
- since we handle national-security sensitive data, and higher ups had the great wisdom to choose a US-owned provider, we now have malicious token minting from our provider in our threat model.
- the admins complain that the old system for issuing permissions was nicer to use
- everyone except me is afraid to touch the OIDC code
- our compile times have suffered
- our testing setup is more complex
even if you don't share my dislike of OIDC, i hope you can at least see why i think working on an OIDC library would be doing free legwork for capital interests, and that there are better things i could use my time and energy on.
it's funny, whenever i write a blog post i try to guess what the most controversial thing will be and i'm always wrong
2/2