Writesas

[JanMarvin]

Rebased the initial work (last commit was in July 2021) on #14 onto current main.
I do not remember anything, therefore the entire approach is HIGHLY EXPERIMENTAL.

• writing mtcars and iris as sas7bdat files works (checked this with SAS studio)
• reading these in readsas works (but a lot of debug information is printed)
• writing a subset from pivottabler::bhmtrains (first 4k rows with trimmed names) works

ToDo

The following is a simple checklist. Marked with * are tasks that are unlikely to be implemented.

• check that writing various cell types works as expected
  • integer
  • numeric
  • date
  • posix
  • character
  • factor (as character)
  • missings
• write multiple pages (iirc readsas currently only supports writing a single page)
• write wide datasets. right now a row must fit on a page and page 1 must contain data. With wide datasets sas splits rows over pages and sas might create a page without data.* (it requires more research and I don't have a use case for this aka I don't need it)
• compression*
• writing 32/64 bit files* (initially I wanted to use 32 bit files because they are much smaller and therefore testing the file format would be easier. unfortunately they did not seem to work)
• files work with current presumably Unicode aware sas. They don't work with latin1 windows and will throw a YZR code bug for unknown reasons. Maybe one of the very many unks defines the encoding.
  I assume that SASYZR is the default compression algorithm used by SAS. This probably is related to case 8, where c8vec contains some kind of column integer information mixed together with a few blanks. Maybe it is possible to disable compression entirely for a file.
• test limits, number of columns/rows* (with a side matrix there's an overflow in some position seeking, which causes 4gb files)
• labels (already working? they only blow up page 1 and I don't have a use case for them)
• various file metadata
• check against SAS
  ...

# in this data frame 485 is almost a full page writing more corrupts the output
dat <- pivottabler::bhmtrains
# dat <- dat[1:485,]

write.sas(dat, "/tmp/trains.sas7bdat")
read.sas("/tmp/trains.sas7bdat")

ETA

Whenever I find time to implement things.