** For FLOAT values, the content is the IEEE754 floating point value in
** native byte-order. This means that FLOAT values will be corrupted when
** database file is moved between big-endian and little-endian machines.
And my immediate reaction is why? SQLite DBs are portable themselves.
Why make the decision to break that portability when using the LSM1
I strongly suspect SQLite4, which also uses LSM, uses a portable encoding
for row data (the contrary would be extremely surprising), so why go for
non-portability on that one datatype, instead of 1) encoding the endianness
in the type,
or 2) forcing a given endianness, or 3) using varints to encode the
mantissa and exponent,
therefore providing the platform portability?