Virtual sites in charmm36-jul2022.ff

GROMACS version: 2021.5
GROMACS modification: NO
Last year I posted a similar question, but this is new problem now.
I followed Andras Wacha suggestions he had given me then and have copied the merged.vsd file to the charmm36-jul2022 release, added the [ constrainttypes ] from the feb2021 rlease to the new ffbonded.itp.

Using gmx 2021.5, pdb2gmx and solvation went well but for the minimization for grompp I have gotten the following errors:
“virtual site MCH1 (Res ACE-23) has non-zero mass 7.5175 (state A) /
7.5175 (state B)
Check your topology.

virtual site MNZ1 (Res LYS-27) has non-zero mass 8.5155 (state A) /
8.5155 (state B)
Check your topology.”

Why dummy atoms MCH1, MNZ1,… got masses, while in the atomtypes.atp Vsite atoms have 0 mass definition?

Interestingly, when I use the charmm36-feb2021 release and the same version of gmx, the virtual site atoms still get the same high masses, but grompp and minimization and simulation go without errors.

I have checked back my older simulations using Vsites with CHARMM36 all-atom force field (November 2018) release and gmx 2016, obviously, had no errors either.