GeezerVison

Fixes #1298.

Summary

DecimalCodec.EncodeDecimal silently corrupted DECFLOAT values whose coefficient or exponent does not fit the target format; only a Debug.Assert checked it. The encoder now brings the value into the format's ranges without changing its numeric value, or throws OverflowException.

Changes

DecimalCodec.FitToFormat (used by EncodeFinite):

• drops trailing zeros of a coefficient that has too many digits (12345678901234567890123456789012340 fits DECFLOAT(34)),
• moves an exponent that is too large into the coefficient (1E384 becomes 1000000000000000E369 in DECFLOAT(16)),
• drops trailing zeros when the exponent is too small (100E-400 becomes 1E-398),
• clamps the exponent of zero,
• throws OverflowException ("cannot be represented exactly as DECFLOAT(n)") when none of that is possible, i.e. when the value would have to be rounded, or overflows or underflows the format.

DecimalCodec is used by both the managed and the native client, so both are fixed.

Tests

• DecimalCodecTests (NoServer) covers, for both formats:
  • representable values, including the format limits,
  • values that are representable only after normalization,
  • values that must throw.
• FbDecFloat16SupportTests and FbDecFloat34SupportTests:
  • server round trip of normalizable values, which were corrupted before,
  • PassesUnrepresentableValueThrows, which also checks that the connection stays usable afterwards.
• Without the fix, 35 of the new tests fail. With it, the embedded + NoServer suite passes: 648 passed, 0 failed.