Re: Writing an Open Access handle that will either read/write to the Db2 file or make a REST call

<Jon>
Because an API call cannot take advantage of RPG's automatic data marshalling features. I often think that this was a major factor in the slow adoption by RPGers in general of web programming. Having to call an API for each and every field that is to be sent to the browser is a very different thing from simply doing (say) EXFMT.

You and I embraced the API approach - to many others it just looked like it was going to be worse than coding assembler or COBOL. Way too much typing.

Your example is frankly a little silly. The handler has to be named on the File definition so the purpose and limitations are somewhat obvious. If the handler is called ReadIFS then the programmer is hardly likely to expect it to eat with CHAIN. Just as they would not expect to use CHAIN with the target file that resulted from CPYTOIMPF that the handler is avoiding the necessity for.

The major reason for defending against unsupported op-codes is that OA does not provide a means by which the handler can identify to the compiler which operations are supported. As a result a maintenance programmer seeing that a file is keyed might be tempted to try to use SETLL if they are not thinking about which handler they are using. The compiler cannot defend against that so the handler must. Or the handler be modified to deal with such requirements if thy make sense in terms of the underlying technology.
</Jon>

@API or not: There is no need to call an API for every field, that is nearly as silly as the internal API interfaces of OA (I don't have so much experience with) or the ARDPGM interface (I do have some experience with). We could provide interfaces very similar to the external interface of OA, something like call someProc(recordType record) bringing back a well defined record - all needed conversion stuff and interpreting the record is done encapsulated in someProc.

@myExample: did you really read it? I was talking about chain and somebody trying setll/read, two ways done with a keyed file, both reading methods - how should one name the file? chainOnlyDoNotSetxxKeyedFile perhaps? BTW: the compiler can't defend against using unsupported OpCodes, as OA implementation is very limited, a little bit like the free format compiler at times it needed nonsens statements like /free /end-free. Minimum would be to me some mechanism like Procedure interface definition to ensure that OA implementation and usage do fit with each other.

The real difrence between your recommendations and mine are:
I do strongly recoomend: Don't use RLA in newly created programms, it's riding a death horse.

Dieter