The exported file cannot be opened by SAS

[Jiang-Li-backup]

I exported data to a SAS file. This file can be read by haven, but not the SAS. I am using SAS EG 7.1.

[hadley]

Please provide a reproducible example.

[Jiang-Li-backup]

Thx. I also compared with the same data set created by SAS. The file generated by SAS EG is twice as big as that generated by R Haven. May I send the data files and code to you by email?

[hadley]

Sure - both haven and SAS versions of the same file would be super helpful.

[Jiang-Li-backup]

I emailed the data and code to you a few days ago. Please let me know if you have not received it. Thx.

[rogerjdeangelis]

r_write_sas7bdat.zip

[rogerjdeangelis]

Observation

Byte 33 has value 33 which indicates that the R SAS datasets use the layout of unix 64 bit.

I will try to read the R SAS dataset on Unix 64.

This should not matter, Win SAS should be able to read the dataset. I will look at the rest of the layout.
Unfortunately, I don't have WPS on Unix so I will create a unix SAS SAS dataset(not WPS) so more of
the R layout can be checked. SAS datasets seemed to have more filler which complicates the comparison.

At least the magic string matches

0-32 bytes Magic string
000000000000000000000000C2EA8160B31411CFBD92080009C7318C181F1011 WPS
000000000000000000000000C2EA8160B31411CFBD92080009C7318C181F1011 R

33-64 bytes
2222003333010232040000000000000000000301181F10112222003333010232 WPS
3300002200010031000000000000000000000000000000000000000000000000 R

[evanmiller]

A quick note about the "filler". SAS files are organized into a "header" followed by 1 or more "pages". By default haven will write a header of size 1024 bytes and pages of size 4096 bytes. This layout was common with older versions of SAS, but newer versions seem to use much larger header and page sizes. The sizes are defined as 32-bit integers at (or near) byte offsets 200 and 204. In any event, extra-large page sizes are the source of filler in the SAS files. If you skip to the beginning of each page it should be easier to analyze the file.

For binary debugging I use hecate.

[rogerjdeangelis]

Thanks for the info and the excellent work.

None of Matt's test datasets have byte 32=33 and byte 35=22. I have checked many other datasets created with different hardware and none of them have the 33/22 combination that write_sas creates

Test datasets for Matt SAS7BDAT

                      32 33 34 35 36 37

compress_no.sas7bdat 22 22 00 32 22 01 02 32 9.0101M3NET_ASRV........
compress_yes.sas7bdat 22 22 00 32 22 01 02 32 9.0101M3NET_ASRV........
lowbwt_i386.sas7bdat 22 22 00 32 22 01 02 32 9.0202M0W32_VSPRO.......
missing_values.sas7bdat 22 22 00 32 22 01 02 32 9.0202M0W32_VSPRO.......
obs_all_perf_1.sas7bdat 22 22 00 32 22 01 02 32 9.0101M3XP_PRO..........
adsl.sas7bdat 22 22 00 33 33 01 02 32 ....9.0202M3X64_ESRV....
eyecarex.sas7bdat 22 22 00 33 22 00 02 31 ....9.0000M0WIN.........
lowbwt_x64.sas7bdat 22 22 00 33 33 01 02 32 ....9.0202M2X64_VSPRO...
natlterr1994.sas7bdat 33 22 00 33 33 00 02 31 ........9.0101M3SunOS...
natlterr2006.sas7bdat 33 22 00 33 33 00 02 31 ........9.0101M3SunOS...
txzips.sas7bdat 33 22 33 22 00 33 33 01 02 31 ........9.0201M0Linux...

[evanmiller]

@rogerjdeangelis Nice detective work!

[realitix]

Hello @hadley, do you have some news about this issue ?