Dataset label truncated after `write_xpt`

[kaz462]

From write_xpt documentation:

Note that although SAS itself supports dataset labels up to 256 characters long, dataset labels in SAS transport files must be <= 40 characters.

The following dataset label in Chinese has 40 characters and was truncated after write_xpt.

(thanks @siye6 for the original example in https://github.com/atorus-research/xportr/pull/194)

label <- "这是一段文字，用来测试在XPTversion5中作为数据集label是否会被截断"
nchar(label, type = "chars")
#> [1] 40
nchar(label, type = "bytes")
#> [1] 88

tmp <- tempfile(fileext = ".xpt")
haven::write_xpt(mtcars, tmp, label = label)
test <- haven::read_xpt(tmp)
attributes(test)$label
#> [1] "这是一段文字，用来测试在XPTv"

nchar(attributes(test)$label, type = "chars")
#> [1] 16
nchar(attributes(test)$label, type = "bytes")
#> [1] 40

^{Created on 2023-12-13 with reprex v2.0.2}

[ynsec37]

Dear developer,

I found that the label length that must be <= 40 is just used for the xpt 5, if the version = 8 the label should be up to 256.

sas xpt version 5

sas xpt version 8

[botsp]

Hi both, May I confirm a question about data conversion. I am trying to convert the .rda file to .sas7bdat, and it seems the "write_xpt" doesn't work as expected. The created sas7bdat file cannot be opened, it always shows "file ... is not a SAS data set".

• write_xpt(my_dataset, "P:/Kevin/CDISC_DataFromR/Pharmaverse/Adam/my_sasdsy.sas7bdat") *

I saw some discussion about this issue and doesn't found a good solution.

1. What is the recommended method for converting the .rda file to a .sas7bdat file?

2. It seems that "write_xpt" works well when converting to an .xpt file. Should I first convert the file to an xpt format and then change it to a .sas7bdat file using SAS? Are there any potential risks associated with this approach?

Looking forward to leanring the insights from your valuable experience. Many thanks!

[ynsec37]

Hi @botsp It seems that write_xpt() may only support the xpt creation.

write_sas() creates sas7bdat files. Unfortunately the SAS file format is complex and undocumented, so write_sas() is unreliable and in most cases SAS will not read files that it produces. write_xpt() writes files in the open SAS transport format, which has limitations but will be reliably read by SAS.

For sas7bdat, I use the same way you mentioned, that is creating the xpt first by R then coverting to sas7bdat by SAS. After converting, I compared results from write_xpt() with SAS datasets directly created by SAS, there is no difference except the variable length.

[botsp]

Thanks for your explanation and this inspire me about the method of sas data conversion. Thank you!

[gorcha]

Hi @kaz462 and @ynsec37,

Thanks for the feedback! This is an issue with our dataset label validation code, and the documentation could be clearer - the dataset label for XPT files is a maximum of 40 bytes rather than characters. Our validation code is currently checking with the default type = "chars" and should be updated to type = "bytes".

@ynsec37 note that the XPT documentation shared above is referring to the variable label length. Although variable labels can be longer in version 8 the maximum dataset label length is still 40 bytes.