[hpoettker]

ReadStat currently writes SAS numeric missing values (., ._, .A–.Z) to .sas7bdat files in a way SAS doesn't understand. The writer puts the inverted ASCII tag character into a standard quiet NaN (0x7FF8…), which isn't the NaN value that SAS actually expects as a basis.

The readstat CLI also drops tagged missing values when writing .sas7bdat: values are never passed to readstat_insert_tagged_missing_value.

Changes

• src/sas/readstat_sas7bdat_write.c: missing values are now encoded the way SAS stores them:

  • The two most significant bytes are 0xFF 0xFF.
  • The third byte is the bitwise complement of the tag code: ._ → 0, . → 1, .A–.Z → 2–27.
  • All other bytes are zero.
  • The byte order follows the machine's endianness.

• src/bin/write/mod_readstat.c: tagged missing values are now written through readstat_insert_tagged_missing_value when the output is .sas7bdat, as already done for .dta.

Verification

• SAS-generated .sas7bdat files containing ., ._ and .A–.Z were converted with readstat in.sas7bdat out.sas7bdat. The output opens in SAS and shows the same missing values as the original. Before this change, none of the missing values survived.