Re: 7.1 and RPG prototypes

[Peter Dow <petercdow@xxxxxxxxx>]

So instead of an error when compiling the program, I'd get an error when 
they try and run the program after it's been moved to production? Better 
for the programmer, worse for the users?

--
*Peter Dow* /
Dow Software Services, Inc.
909 793-9050
petercdow@xxxxxxxxx <mailto:petercdow@xxxxxxxxx>
pdow@xxxxxxxxxxxxxx <mailto:pdow@xxxxxxxxxxxxxx> /


On 3/5/2018 9:08 AM, D*B wrote:
> Hi,
> I don't realy understand, why it should be so hard to do.
> Right now we have:
> at compiletime: checking against the declaration of a prototype, 
> rather weak hopefully the same prototype is used as in the 
> implementation.
> at bind process: resolving symbols by name, binding by position (could 
> be stronger, if bound by name), creating a signature (could be 
> stronger if the signature would use name and parms) - the real problem 
> here is, that many people fake the signature to avoid rebind needs 
> (strange, very strange!!!)
> at runtime: signature check.
>
> IMHO it wouldn't be too hard ton enhance this process:
> - no check needed at compiletime, just making a list of unresolved 
> references, including parms would be enough
> - at bind time: check use of unresolved references against the list of 
> exports in all components (modules, SRVPGMS) available by parm of 
> bindprocess.
> in other words: move the checking usage against PR from compile to 
> bind. To make this happen, write the declarations of the exports in 
> the header section of the Modules/SRVPGMs. This seems to me very 
> similar to putting the source into the object for debug purpose.
>
> Regards
>
> Dieter
>
> <Jon>
> OK - I see where you are coming from - but, as you seem to realize, 
> what you would like is hardly a practical option.
>
> What would be required to enable the "PI rules" philosophy is a system 
> wide extension that captures the full details of the parameters and 
> return value. Right now the closest we have to that is PCML and that 
> has multiple problems including an inability to be able to represent a 
> number of parameter types. Even if that problem were addressed (not 
> likely worth IBM's investment) then there is still the problem of late 
> binding. For instance a dynamically called program can be subject to 
> library list and other factors - how can the compiler validate against 
> something that may not even exist at compile time. Same problem with 
> Service Program procedures that are dynamically bound - or bound via 
> procedure pointer.
>
> I am interested in minimizing problems with what we have and asking 
> for extensions (such as the RFE in question) that help to avoid 
> errors. I think we have more chance of IBM i being open sourced than 
> seeing the types of changes needed to avoid protos altogether.
>
>
> Jon Paris
> </Jon>