Because of a misguided attempt to protect users from doing what users want to do, JMRI versions 4.9.2 through 4.17.1 attempted to "normalize" (a.k.a. change the system names of things, like Memory objects, that users create) to prevent user from making "mistakes". This normalization was deliberately silent (i.e., you never got any feedback that it happened). In JMRI 4.17.2, all normalization of system names that could be identified was ripped out, so versions of JMRI 4.17.2 and newer do not protect you from using the system names you want (JMRI 4.17.2 and newer will validate system names against protocol requirements, and should notify you of invalid names in that case).
That silent normalization process would conflate IMTRAIN01 and IMTRAIN1 into a single object, explaining why it was possible to create those in JMRI versions 4.9.1 or earlier, and get no error working with them in 4.9.2 through 4.17.1.
It is clear we have found a bug in JMRI where some normalization of system names is still occurring, but should not.
