Re: IBM i "Open I/O" Architecture

[CRPence <CRPbottle@xxxxxxxxx>]

  FWiW [and what you probably already know, but others probably do 
not], that alluded override invocation [redirecting to a program] is 
already quite similar to how the OS already functions.

  The /common data management open/ [QDMCOPEN] currently processes 
the overrides and establishes which program is to be called for each 
of the supported I/O methods according to the type [e.g. device & 
access path] of the open; established in the Open Data Path created 
for the OPEN request.  Defined as just one program however, the 
methods might best be a feature of the runtime I/O versus open; i.e. 
the defined program restricts the unsupported method by an error 
versus protecting the access via the common data management layer.?

  There are feedback areas for the open itself [record length, 
...], keys, positioning, errors, etc.  All that formalized and 
exposed to a user-defined program, such a program could then provide 
access to the data [stream] in any container that is not already a 
supported *FILE type, via any HLL program coded to use the builtin 
I/O operations for that HLL; e.g. OPEN, EXFMT, WRITE, READ, READE, 
RCVF, SNDRCVF.

  Not to imply anything of the subject support, as I know nothing 
of it.  However as a true architecture [being "open" is unnecessary, 
only "published" is required], any language on any system could add 
statements to utilize the architected open\close, I/O operations, & 
feedback, in order to take advantage of anything coded to handle 
those requests also according to the architecture.  With such an 
architecture, the RPG [both the language for compiling, and existing 
sources] on i, could be more easily ported to AIX if someone decided 
to build the effective device and database file support against 
which those file related opcodes perform.  The same could be done 
without anything actually being /open/, but how much and how soon 
would anything get done if entirely dependent on IBM to make it happen?

Regards, Chuck

Mark S. Waterbury wrote:
> Why is this idea being presented as only an RPG IV enhancement?
> I think it would be far better to provide this capability in
> the "base" OS, namely, via a mechanism such as the "override"
> commands, so that you could do something like this:
>
> OVRDBF   FILE(MYFILE)   PGM(MYLIB/MYPGM)
>
> This would then tell the system to call the named program for all
>  file-related activity for this file (open, read, write, close,
> etc.). Of course, a suitable parameter list structure would need
> to be defined for such an "exit" program, but conceptually this
> should be able to work similar to triggers.  This would work in a
> manner similar to RPG's SPECIAL files, but would work for all
> opcodes, including EXFMT.
>
> I think this facility should be supported on all of the OVRxxx
> commands (OVRDBF, OVRDSPF, OVRPRTF, etc.) -- so that it would be
> possible to "redirect" any existing file to another device (or
> the web).
>
> This kind of approach would enable such an enhancement to work
> with any supported language or compiler.
>
> What do you all think?