Yun (Julie) Zhuo
, PRA Health Sciences
In today's integrated and interconnected world, sharing SAS data among organizations and countries has become increasingly common. The transcoding error, which may occur during data transfers between incompatible SAS encoding environments, remains one of the most mysterious SAS errors to programmers. This presentation discusses three steps to demystify the transcoding issues. First, encoding, transcoding and other background information will be explained. Second, SAS procedures, options and other techniques that are useful for troubleshooting will be given. Finally, the character variable padding (CVP) engine and SAS configuration files are provided as solutions to the problem. By the end of the presentation, the audience should have a clear vision for troubleshooting and solving transcoding errors.
Watch Transcoding: Understand, Troubleshoot, and Resolve a Most Mysterious SAS® Error on the SAS Users YouTube channel.
If we live in a perfectly standardized world, sharing data would be a smooth copy-and-paste from computer to computer. Unfortunately, we often receive data from an incompatible computing environment with an incompatible session encoding. Problems thus arise.
SAS Cross-Environment Data Access (CEDA) is designed to tackle the problem of incompatible encodings. However, it has limitations. For example, data loss or truncation may occur when the data contain extended ASCII or other special characters. When this happens, CEDA will issue a message such as the following:
ERROR: Some character data was lost during transcoding in the dataset
SOURCE.DM. Either the data contains characters that are not representable
in the new encoding or truncation occurred during transcoding.
What can we do when confronted with such a mysterious error message? Pushing back the data to the data provider is not always a feasible option. In this paper, we aim to demystify the error message, and provide tools, workarounds, and solutions to help resolve the error at the data recipient’s end.
Before we attempt to troubleshoot and resolve the transcoding error, it is helpful to understand why the error occurs. In this section, we are going to provide background information to help demystify the error message.
In computers, character data is stored as a series of bytes, which is made up of binary digits called bits. A coded character set associates each character with a number. Computer encoding maps the bits and bytes of stored data to the correct characters via the coded character set. Since our computing environments are not perfectly standardized, there are different types of encodings in use, and each encoding has its own character set.
|Name of Encoding||Associated Character Set||Distinguishing Factors|
|WLATIN1||ASCII and Extended ASCII character set||Represent 256 characters. Each character is stored in a single byte.
Single Byte Character Set (SBCS)
||ASCII, Katakana, and other Japanese characters||Characters are stored in either 1 or 2 bytes.
Double Byte Character Set (DBCS)
|UTF-8||Unicode character set including ASCII, foreign languages, special symbols, and more||Represent over 120,000 characters. Each character is stored in 1 to 4 bytes.
Multi-Byte Character Set (MBCS)
Table 1. Examples of Commonly Used Encodings
Table 1 gives examples of some commonly used encodings and their associated character sets. The last column of the table shows two distinguishing factors of different data encoding types:
1. The number of characters each encoding can represent differs. While the WLATIN1 encoding, the most widely used encoding type in the Western countries, can only represent 256 lower ASCII, ASCII control, and extended ASCII characters, the UTF-8 encoding, a common encoding form of the Unicode standard, can represent over 120,000 characters. Figure 1 shows all the lower ASCII and extended/upper ASCII characters.
Figure 1. ASCII and Extended ASCII Characters
2. The storage size of each encoding differs. While the single byte character set (SBCS) only takes up one single byte for each character, the multiple byte character set (MBCS) needs up to four bytes to store characters. For example, Figure 2 shows that in order to represent the extended ASCII character ‘á’, the WLATIN1 encoding needs only one byte, but the UTF-8 encoding uses two bytes to represent the same character.
Figure 2. Difference in Storage Size for WLATIN1 and UTF-8 Encodings
Now what is transcoding? Transcoding is the process of converting from one computer encoding to another. In SAS, each programming session has a pre-configured encoding. If the encoding of the input data differs from the encoding of our currently executing SAS session, SAS will invoke the Cross-Environment Data Access (CEDA) engine to transcode the data, automatically converting the input data encoding to the encoding of the SAS session in the target computer. Figure 3 demonstrates the process of transcoding from one computing environment to another.
Figure 3. Cross-Environment Computer Transcoding Process
When SAS invokes CEDA and transcoding occurs, SAS issues the following informational note to the log:
NOTE: Data file is in a format that is native to another host, or the
file encoding does not match the session encoding. Cross Environment
Data Access will be used, which might require additional CPU
resources and might reduce performance.
Unfortunately, the SAS CEDA does not work as miraculously as we hope. Transcoding error occurs when the CEDA transcoding process fails. The transcoding error contains two components, as shown below in Figure 4, each pointing to a potential problem that causes the error.
Figure 4. Two Components of the Transcoding Error Message
The first problem is characters that are not representable in the new encoding. As we explained earlier, some encodings such as the UTF-8 encoding store more than 120,000 characters covering 129 scripts and multiple symbol sets. ASCII characters, Asian languages and non-ASCII special characters are all represented in the UTF-8 encoding. By contrast, the WLATIN1 encoding only contains 256 ASCII and extended ASCII characters. Any character that is not included in the WLATIN1 encoding will be lost if we try to transcode from UTF-8 to WLATIN1. In short, the unrepresentable characters error occurs when the input data set contains characters that do not exist in the target computer encoding, triggering the transcoding error message in the log.
The second problem is character data truncation. As we explained earlier, one distinguishing factor between encodings is the storage size. The single byte character set (SBCS) only takes up one single byte for each character, but the multiple byte character set (MBCS) may need more than one bytes to store characters. Therefore, if the target data set uses a MBCS such as the UTF-8, the length of the character variables defined in the source or input data set may not be big enough to hold all the characters for storage in the target encoding environment. When this occurs, SAS will trigger the transcoding error message to warn you about truncation.
Now that we understand how transcoding error occurs, the next step is to troubleshoot the issues. There are a number of SAS options and procedures that are helpful. In this section, we will provide answers to specific questions encountered during troubleshooting.
The first question we ask is which of the two components in the error message has caused the problem. Is it unrepresentable characters or data truncation? To answer this question, we need to find out the encodings used in both the source data and the currently executing SAS session. If SAS transcodes SBCS (e.g., WLATIN1) to MBCS (e.g., UTF-8), then the problem is data truncation. On the other hand, if SAS transcodes a large character set to a smaller character set, then the problem is most likely the unrepresentable characters.
We can find out the encoding of the input data using one of the following two methods.
The first method is the CONTENTS procedure. That is, we run the following simple SAS CONTENTS procedure:
proc contents data=source.DM; run;
Then we look for the encoding row in the output:
Figure 5. The Encoding Attribute in the PROC CONTENTS Output
The example in Figure 5 shows that the input data set SOURCE.DM was created under the Windows computing environment using the WLATIN1 encoding. If the transcoding error occurs and the currently executing SAS session uses a different encoding with higher byte character set (e.g. UTF-8), then the problem is data truncation.
If you see the following data representation and encoding values instead in the PROC CONTENTS output:
then the input data set was created under the UTF-8 encoding. If the transcoding error occurs and the currently executing SAS session uses a different encoding with a smaller character set (e.g. WLATIN1), then the problem is unrepresentable characters.
The alternative method to find out the encoding of the input data is the ATTRC function.
After we run the following SAS codes:
%let dsn=source.DM; %let dsid=%sysfunc(open(&dsn,i)); %put &dsn ENCODING is: %sysfunc(attrc(&dsid,encoding)); %let rc=%sysfunc(close(&dsid));
The encoding will be printed to the SAS log as shown below:
We can find out the encoding of our currently executing SAS session using one of the following two methods.
The first method is the OPTIONS procedure. That is, we run the following simple SAS OPTIONS procedure in our current SAS session:
proc options option=encoding; run;
Then we look for the encoding information printed to the SAS log.
Figure 6. The Encoding Attribute in the SAS Log from the OPTIONS Procedure
The example in Figure 6 shows that the current SAS session in our computer uses the WLATIN1 encoding to represent character data.
Alternatively, we can use the GETOPTION function. That is, we run the following SAS code in our current SAS session:
We will find the similar encoding information printed to the SAS log. As shown below:
The above examples show that the current SAS session uses WLATIN1, the encoding with single byte character set. If transcoding error occurs and the input data has a larger character set, then the problem is the unrepresentable characters.
Unrepresentable characters are often buried under mountains of data on hand. A single unrepresentable character in an inconvenient location can stop SAS from transcoding. It is helpful if we can identify the unrepresentable characters causing transcoding issues and visually inspect them before we make decisions on our next move.
In the example below, shown in Figure 7, input data set has the UTF-8 encoding but the current SAS session in our computer uses the WLATIN1 encoding. We use a simple data step to read the input data set. Not surprisingly, a transcoding error occurs. In the log, in addition to the transcoding error message, SAS also prints a warning message indicating that the data step stopped right after it finished processing 799 observation. What it means is that on observation number 800, there must be some unrepresentable characters that stopped the data step from processing.
Figure 7. Warning Message from Data Step Shows Where Error Occurs in Data
To confirm our theory, we use the ENCODING=ASCIIANY SAS option to bypass the transcoding error in order to take a visual inspection of the row number 800 in the input data set. Note that the ENCODING=ASCIIANY option will allow SAS to ignore the UTF-8 encoding altogether. SAS will interpret every bite in the input data using its single byte ASCII coded values.
The syntax is the simple SAS codes shown below:
Data indata.cm4; Set rawdata.cm4 (encoding=asciiany); Run;
After we execute the above codes, we will be able to read the entire input data set with 1163 observations without any error or warning messages, as shown below in Figure 8.
Figure 8. Bypass Transcoding Error with the SAS Encoding=ASCIIANY Option
Then, to locate the unrepresentable characters, we open the output data set we have created without error, scroll down to row number 800, and then as shown in Figure 9, we immediately spot some weird characters that do not make any sense.
Figure 9. Locate the Unrepresentable Characters in the Output Data Set
What happened is that, with the ENCODING=ASCIIANY bypassing option, SAS will interpret every bit in the input data using its single byte ASCII coded values. It is most likely that the unrepresentable characters will be turned into ASCII special characters such as those in the blue rectangle in Figure 1. ASCII and Extended ASCII Characters.
Therefore, the ENCODING=ASCIIANY SAS option provides a way for us to locate the unrepresentable characters, and visually inspect the data. It also helps us estimate the extent of our issues. However, it does not completely resolve our problem.
The transcoding error has two components. Each component points to a potential reason for the transcoding failure: data truncation or unrepresentable characters. Each component also has its own solutions. If the error is caused by data truncation, we can use the Character Variable Padding (CVP) engine to avoid truncation. If the error is caused by unrepresentable characters, we propose updating the session encoding in the SAS configuration file if easy walk-around solutions do not work for you.
If you find out that the encoding of your session uses the MBCS but the encoding of the source data set uses the SBCS or the DBCS, the error message you encountered above usually means that there is not enough space in one or more character columns to convert the data to the target encoding. If this is the case, you should consider using the Character Variable Padding (CVP) engine to avoid truncation.
What is the CVP engine? It is a read-only engine that expands the length of the character variables at your request. It serves as an intermediate engine that is used to prepare the data for transcoding. By default, it multiplies the character variable lengths in the input data set by 1.5. After the lengths are increased, the primary engine is used to do the actual processing.
In the following example, the input data set was created under the Windows computing environment using the WLATIN encoding which represents a SBCS character set. As shown in Figure 10, the input data set contains one character variable and two records. Both records contain four ASCII and extended ASCII characters. The length of the character variable is set to 4, which provides enough space in the WLATIN encoding environment to store the data because SBCS only takes up one single byte for each character.
Figure 10. Example Input Data Set with Extended ASCII Characters
When we try to read the above input data set into a different encoding session such as the UTF-8 encoding SAS session. The character variable will be truncated. Because in the UTF-8 encoding, it takes more than one byte to store special characters such as the ‘Ø’ and ‘Ü’ in this example.
To resolve this issue, we run the following code to invoke the CVP engine with a default multiplier of 1.5:
libname source cvp 'Source-data-library';
After running the above code, the length of the variable will be automatically increased from 4 to 6, as demonstrated below in Figure 11.
Figure 11. Variable Length Before and After CVP Engine Processing
In order to multiply the character variable lengths by a different amount, you need to use the CVPMULTIPLIER= option to specify a number. You can specify a value from 1 to 5, or you can specify a value of 0 to let the CVP engine determine the amount. Here is the example code that uses a user-specified multiplier:
libname source cvp 'Source-data-library' cvpmultiplier=2.0;
It may take some experimenting to determine the correct multiplier in order to make sure you have adequate space for all situations. During the experimenting process, be sure to review the log carefully and watch for truncation error messages.
Since the CVP engine is read-only, an additional LIBNAME statement is required in order to save a permanent copy of the converted data set. Here is the full example code:
libname source cvp 'Source-data-library'; libname target 'Target-data-library'; proc copy noclone in=source out=target; run;
In the above example, the NOCLONE option in PROC COPY is required. The option makes sure SAS adopt the target operating system data representation, the target session encoding, and other relevant attributes.
To conclude, we run the risk of truncating character data when transcoding into a character set that stores data in multiple bytes. By expanding the length of the character variables,
the CVP read-only LIBNAME engine is an effective and elegant solution to avoid data truncation.
If you find out that the source data encoding is UTF-8 but our SAS session encoding is WLATIN1 or any encoding other than the comprehensive UTF encoding, the problem is most likely the unrepresentable characters because some characters in the UTF-8 character set do not exist in other encoding types. Although the CVP engine resolves the issue of truncation, it is not helpful if the transcoding error was triggered by the characters that are not representable in the target encoding in our SAS session.
What can we do? The first impulse is to contact the data provider and ask them to remove or replace all the unrepresentable characters. However, this will involve negotiations with outside organizations, and it may not be feasible or convenient in many instances.
The solution is to update the encoding type in our currently executing SAS session. We can update the encoding type to either the UTF-8 encoding or the same encoding of the source data so that all characters in the source data can be represented in the SAS session in our computer.
The encoding type of our SAS session is configured in the SAS configuration file. Figure 12 below is a partial snapshot of an example SAS configuration file that defines the encoding of the SAS session as UTF-8.
Figure 12. Encoding Definition in the SAS Configuration File
To update the SAS session encoding, firstly, we need to locate the SAS configuration file in our computer. After running the following OPTIONS procedure:
Proc options option=config; Run;
The location of the SAS configuration file will be printed to the log, as show below in Figure 13.
Figure 13. Locate the SAS Configuration File using the OPTIONS Procedure
By default, during SAS installation, SAS installs three configuration files to our computer. Figure 14 shows three default sub-folders and each sub-folder contains a copy of the SAS configuration file with a specific type of encoding.
Subfolder ‘1d’ contains a copy of the configuration file with the encoding type of SHIFT-JIS.
Subfolder ‘en’ contains a copy of the configuration file with the encoding type of WLATIN1.
Subfolder ‘u8’ contains a copy of the configuration file with the encoding type of UTF-8.
More encoding types could be installed through a customized SAS installation.
Figure 14. Default Locations for SAS Configuration Files
The second step is to instruct SAS to use a copy of the SAS configuration file that has the desired type of encoding: the UTF-8 encoding or the encoding used in the data.
If you use the Windows platform, you can customize your shortcut icon to instruct SAS to invoke UTF-8 by following the steps below, as illustrated in Figure 15.
1. Locate your SAS shortcut icon or your SAS Enterprise Guide shortcut icon. If you do not have a SAS shortcut icon already created, you have the option of creating one before customizing it following the below steps.
2. Right-click the shortcut icon and select Properties.
3. On the Shortcut tab, in the Target line, update to instruct SAS to use the UTF-8 version of the NLS (National Language Support) configuration file (sasv9.cfg). For example:
"C:\Program Files\SASHome\SASFoundation\9.4\sas.exe" -CONFIG "C:\Program Files\SASHome\SASFoundation\9.4\nls\u8\SASV9.CFG"
4. Click 'OK'.
5. Optionally, you can rename the shortcut icon to indicate this shortcut will invoke SAS with the UTF-8 encoding. On your desktop, you can also create additional shortcut icons that have other types of encoding.
Figure 15. Update SAS Session Encoding through the SAS Shortcut Icon
If you invoke SAS through a command line or a batch file, you have the option to specify the configuration file location through the CONFIG system option. For example, the code below invokes SAS using the UTF-8 encoding version of the configuration file:
C:\Progra~1\SASHome\SASFou~1\9.4\sas.exe -sysin ProgramName.sas -config C:\Progra~1\SASHome\SASFou~1\9.4\nls\u8\SASV9.CFG
For detailed instruction on invoking SAS with different language and encoding sessions on the Windows platform, refer to the SAS Technical Paper: Setting up SAS9 National Language Support in Microsoft Windows Operating Environments.
If you use the UNIX platform, SAS is invoked by Bourne Shell scripts. The invocation scripts are named using the language codes of the installed language. For example, sas_en invokes the English version, and sas_u8 invokes the Unicode version. Follow the steps below to update your session encoding:
1. Locate the invocation script under the following directory:
2. Update the script to change the SAS session locale or encoding.
For detailed instruction on invoking SAS with different encoding sessions on the UNIX platform, refer to SAS Configuration Guide for SAS 9.4 Foundation for UNIX Environments.
To conclude, the encoding of our SAS session may not be able to accommodate special characters in the source data set. After we convert our SAS session encoding to either UTF-8 or a compatible type of encoding, we will be able to avoid the transcoding error. We will be able to work with all special characters in the data including foreign languages and other non-ASCII characters.
We may encounter the mysterious transcoding error when we try to work with data from a different encoding environment. This paper shows that if we understand computer encoding and how the transcoding occurs, the problem is not as mysterious as it appears to be.
There are two potential reasons for the transcoding failure. If the reason is data truncation, we can use the CVP engine to expand the length of the character variables in the data. On the other hand, if the reason is unrepresentable characters, we may have to convert our own SAS session encoding to be compatible with the encoding in the data. Either way, data recipients will be able to resolve the problem at their end.
Bouedo, M. 2020. “The SAS Encoding Journey: A Byte at a Time.” Proceedings of the SAS Global Forum 2020 Conference, Cary, NC: SAS Institute Inc. Available at https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/4561-2020.pdf
Carlton, J. 2017. “SAS Blogs: Demystifying and Resolving Common Transcoding Problems.” Accessed March 6, 2018. https://blogs.sas.com/content/sgf/2017/05/19/demystifying-and-resolving-common-transcoding-problems/
Dutton, D. 2015. “Data Encoding: All Characters for All Countries.” Proceedings of the PharmaSUG 2015 Conference, Cary, NC: SAS Institute Inc. Available at https://www.lexjansen.com/phuse/2015/dh/DH03.pdf
Li, H. 2020. “Turn Yourself into a SAS® Internationalization Detective in One Day: A Macro for Analyzing SAS Data of Any Encoding.” Proceedings of the SAS Global Forum 2020 Conference, Cary, NC: SAS Institute Inc. Available at https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/4645-2020.pdf
SAS Institute Inc. 2014. Usage Note 52716. Cary, NC: SAS Institute Inc. Available at http://support.sas.com/kb/52/716.html
SAS Institute Inc. 2015. Setting up SAS® 9 National Language Support in Microsoft Windows Operating Environments. Cary, NC: SAS Institute Inc. Available at http://support.sas.com/techsup/technote/ts801.pdf
SAS Institute Inc. 2016. Moving and Accessing SAS® 9.4 Files, Third Edition. Cary, NC: SAS Institute Inc. Available at http://documentation.sas.com/?docsetId=movefile&docsetTarget=titlepage.htm&docsetVersion=9.4&locale=...
SAS Institute Inc. 2017. Configuration Guide for SAS® 9.4 Foundation for UNIX Environments. Cary, NC: SAS Institute Inc. Available at
SAS Institute Inc. 2017. Migrating Data to UTF-8 for SAS® ViyaTM 3.3. Cary, NC: SAS Institute Inc. Available at http://documentation.sas.com/?cdcId=vdmmlcdc&cdcVersion=8.1&docsetId=viyadatamig&docsetTarget=titlep...
SAS Institute Inc. SAS Technical Paper: Multilingual Computing with SAS® 9.4. Cary, NC: SAS Institute Inc. Available at https://support.sas.com/resources/papers/Multilingual_Computing_with_SAS_94.pdf
Stackhouse, M. 2018. “UTF What? A Guide for Handling SAS Transcoding Errors with UTF-8 Encoded Data.” Proceedings of the PharmaSUG 2018 Conference, Cary, NC: SAS Institute Inc. Available at https://www.pharmasug.org/proceedings/2018/BB/PharmaSUG-2018-BB08.pdf
Zhuo, Y. 2018. “Tips and Fixes for Cross-Environment Batch Transfer of SAS® Data.” Proceedings of the PharmaSUG 2018 Conference, Cary, NC: SAS Institute Inc. Available at https://www.pharmasug.org/proceedings/2018/BB/PharmaSUG-2018-BB14.pdf
Registration is open! SAS is returning to Vegas for an AI and analytics experience like no other! Whether you're an executive, manager, end user or SAS partner, SAS Innovate is designed for everyone on your team. Register for just $495 by 12/31/2023.
If you are interested in speaking, there is still time to submit a session idea. More details are posted on the website.