expr in current HEAD returns the following on Motorola M68K architecture:
% expr 0.1
% expr 1e-10
The bug is triggered only when the final result is shown, and only when the resulting number is less than one in absolute value. Then expr replaces mantissa by zero.
% expr (1.0/10)
% expr (1.0/10)*10
Also, expr.test fails with the following:
Test file error: unable to alloc 3988184000 bytes
(I don't know if it's the same bug or another one).
8.4.16 works fine and passes all the tests.