Do you mean:

perl -e 'print 5.1 + 100005.2 - 100005.2'
5.09999999997672?
[download]

Note that this does what you expect:

perl -e 'print 5.1 + ( 100005.2 -100005.2 )'           [12:36pm]
5.1
[download]

To summarize, the bible , most rational numbers are not representable exactly by binary numbers. When the larger constant 100005.2 is added to the smaller constant 5.1 the representational errors in the larger constant can swamp out the smaller constant.

Re-arranging terms as I did in the second example can help. A more universal method is to insure that all constants have exact representations base 2. i.e.

%perl -e ' printf "%f  %f \n", 0x186A5, ( 5.1 + 0x186A5 ) - 0x186A5'
100005.000000  5.100000
[download]