[forgot_old_user]

But wouldn't the same issue apply to user-space TCP implementations too? User-space TCP implementations too could have "path-dependent sequence of accidents" which a power user might eventually need to figure out?

[jeffbee]

Yes but instead of being a 35-year-old accretion of mistakes, a user-space network stack is likely to be part of a more typical software lifecycle, that gets updated more easily and ultimately replaced. Also such things are dramatically easier to debug.

[loeg]

Happily Linux is at most a 32-year old accretion of mistakes (and I'm not completely confident 0.01 even had TCP).