Hi Bastiaan,
I successfully developed the pipeline using the first approach and provided topup outputs as one of the -eddy_options. I don’t think there may be any differences in the results using either approach. Regarding your question, please check this issue which may explain it: How to pad a slice to data for topup
As far as I know, dwifslpreproc will pad the data before topup if your data has odd dimensions. So I guess Synb0-DISCO cannot handle that, so you can either go with the 2nd approach or pad your raw dMRI before using Synb0-DISCO.
Best,
Amir