No, it's fine -- cuz what I meant by hitting the nail askance -- is that eval really didn't like binary embedded in the source it was 'eval'ing -- as you mentioned. That section was being 'eval'ed twice, once by the new code that looped through the table to create the packages, and second, in the 'use pkg X.Y.Z;' code because you can't use an expression with 'use', so the only way to put in the version I wanted was have the eval "evaluate" the expression so it would behave normally.

My "hints" were that in some of the version strings I had quoted them and had done so deliberately, while in some of the I had not done so (again deliberately) -- and that the final test resulted in some tests that were quoted strings, while others used the 'binary' (as formed by the vector) value of the version string.

That's when I put it together -- why didn't the binary cause problems in original code -- because it is only turned into binary as it is eval'ed in the test case. Since I was evaling it twice, my first eval had to result in both a quoted (the qw('quoted string')) and a non-quoted value (qw(vec.tor)) for the different test cases.