This document covers the following topics:
Prior to Adabas version 7, converting data for Adabas buffers between different machine architectures (ASCII, EBCDIC) was handled by Entire Net-Work. With the increasing use of applications where clients and servers (that is, the databases) have different encodings, it has become necessary to expand the data transfer and conversion capabilities of Adabas itself. To this end, Entire Net-Work determines whether the target database has translation capabilities, and if so, passes the unconverted data on to the database for conversion there.
An additional advantage of translating data within Adabas is that other transport mechanisms can now be supported. For UES-enabled databases, Adabas version 7 supports Entire Net-Work access to the OS/390 or z/OS mainframe database through the TCP/IP protocol from web-based applications or from PC-based applications such as Software AG's Jadabas. See the ADARUN parameters TCP and TCPURL in the Adabas Operations documentation for more information.
Adabas data translation occurs as follows:
The client application can specify a special encoding and communicate it to the Adabas nucleus at session open (OP command).
The LNKUES/ADALNK converts Adabas buffer data depending on the architecture of the caller.
A number of utilities provide for special encoding and architecture settings.
EBCDIC to ASCII and ASCII to EBCDIC translation tables are located in the Adabas Installation documentation. A table listing the encoding keys provided with Adabas version 7 is located in the Adabas Command Reference documentation and the Adabas Utilities documentation.
Adabas recognizes four types of encodings that can be specified in parallel:
| Encoding | Character string encoding ... |
|---|---|
| File | is stored and processed internally |
| Default User | is used as the default for Adabas local call interface requests and for ADACMP DDEBAND. |
| User | overrides the "default user" encoding for a user session or an ADACMP execution. This is used to adapt to the special needs of a client program. |
| Collation | is in acceptable sort order. The collation can be defined by language and country standards commonly identified with a "locale" definition. |
Since user data does not require conversion, Adabas equates the local "default user" and "file" encoding to increase processing speed. Remote requests with ASCII architecture are converted using the database "default ASCII user" encoding.
Double-byte character sets are converted using the native mainframe EBCDIC architecture encoding: host DBCS from IBM, Fujitsu Ltd., or Hitachi.
Special applications or remote clients select a specific "user" encoding that fits their processing environment at session open.
To ensure round-trip compatibility between architectures and encodings, Adabas uses a file encoding that holds the superset of all characters defined in the "default user" and any specific "user" encodings. For wide-character fields, such a file encoding defaults to the universal character set encoding Unicode.
Collation encoding is defined for a descriptor field. Values for this encoding are obtained algorithmically by calling a collation exit programmed to produce culturally correct sorted keys; that is, a character dictionary. Collation encoding may be defined for both alphanumeric and wide-character fields; the collation encoding/exit is defined on the file level for alpha and/or wide descriptor fields.
With ADABAS 7.4.2 the new conversion exit ADACOX is available. ADACOX supports context-sensitive conversion between Windows-1256 and EBCIDIC Arabic or EBCDIC Farsi code page with UES enabled databases.
While Arabic characters are unshaped in Windows-1256, the supported EBCDIC encodings use shaped forms depending on previous or following characters. In addition, for certain consecutive characters the combined form is used e.g. LAM-ALEF ligature.
Currently, no support is included for conversion from logical to visual order and vice versa, or symmetric character conversion.
The conversion exit will always be loaded for UES enabled databases.
When a new conversion between two encodings is first used, the exit is queried whether it supports the conversion. If it does, the exit will be called for any such conversion; if it does not, Adabas and/or Entire Conversion Services will do the conversion.
However, the conversions defined in ADACOX need to be backed by corresponding ECS objects; for example, for the conversion 420 to 1256 the character set properties are determined by the ECS objects.
When performing searches using the S operator, the high value is usually a sequence of X'FF' bytes.
With UES=YES, the source and target code pages control the conversion of data. The conversion of the character X'FF' depends on its mapping to the target code page. It is therefore possible that the X'FF' will not remain the X'FF' in the converted value.
For example, when converting from 819 (ISO 8859-1 Latin1) to 37 (EBCDIC), the Latin small letter 'y' with diaeresis is mapped from X'FF' to X'DF'. As a result, searches find fewer or even no records.
Adabas version 7.4 solves this problem as follows: with UES=YES and Alpha (or Wide) conversion, all FROM-TO Search/Logical Read Criteria are handled in such a way that in the TO criterion the high value characters at the value end:
are preserved when converted into the internal search value, and
are excluded from value conversion.
Note:
This solution is not implemented for the value operators (EQ, GT, GE,
LE, and LT). It is limited to the TO value of FROM-TO search criteria (S
operator). This applies to alpha and wide-format fields and to the Alpha/Wide
format parts of Super and Sub Descriptors.
The following restrictions of Entire Net-Work are continued with Adabas translation:
compressed records (FB=C) are not converted.
text literals are not converted and are passed as is. When reading records, a literal is returned unchanged (for example, FB AA, `-do not convert-',BB).
prefetch option P is not supported in conversion.
ET data is not converted. When reading, ET data is padded with EBCDIC blanks.
Additional restrictions imposed by Adabas include the following:
for all C`Xn' command codes used by CSCI, only the control block is converted; not the buffers. This applies only for Adabas version 7 servers.
Entire System Server (NPR) / XCOM applications are not included in the scope of Adabas translation. Those applications need to do their own translation.
OS/2 unpacked numbers sign X`Dn' is not used and therefore, is not supported.
Adabas does not provide user translation exits for field-level translation. Such exits are provided by Entire Net-Work.
Although differences between Adabas versions running on various platforms are gradually being reduced, the following considerations apply when porting applications:
| Mainframe | Open Systems | OpenVMS | |
|---|---|---|---|
| Fixpoint field length | 2 and 4 only | 1,2,4,8 | 1,2,4,8 |
| Binary superdescriptor format default is U (unpacked) | No | Yes | Yes |
| Signed binary superdescriptor | Yes | No | No |
| Binary superdescriptor format conversion | Yes | No | No |
| Superdescriptor with MU and PE fields | Yes | No | No |
| Superfields and subfields | Yes | No | No |
| Superdescriptor with floating-point format parents | No | Yes | Yes |
| Maximum length of unpacked fields | 29 | 29 | 27 |
| Maximum length of packed fields | 15 | 15 | 14 |
| Prefetch option for read and ET/BT commands | Yes | No | No |
| Long alpha (LA) field option | Yes | No | No |
| Field arithmetic update option in format buffer | No | Yes | Yes |
| MC command | No | Yes | Yes |
| Hyperfield value generation from value buffer | No | Yes | Yes |
Additionally, user data provided in mainframe ADALNK user exits is not sent to ASCII machines.
