NEWS.md
write.por() no longer fails on negative numbers. The sign was dropped before the number was encoded, which produced an unreadable value and on some platforms crashed the session (#36). Files also carry two more digits of precision than before.
read.por() no longer loses precision on values with long mantissas. The digits were accumulated into a single number that grew past what a double can hold exactly, so the low bits were lost before the value was assembled. 1e6 used to read back slightly wrong; it is now exact.
Portable files still do not round-trip bit-for-bit: about 75% of arbitrary doubles survive a write and read unchanged, the rest differ in the last decimal digit. That is far below any practical significance, but exact comparisons such as identical() will fail on some values.
These changes have been checked against readspss itself. They have not been checked against SPSS.