|
From: | Jeff Chimene |
Subject: | Re: [open-cobol-list] Proposed addition to OpenCOBOL: an object module generator |
Date: | Mon, 25 May 2009 14:21:32 -0700 |
User-agent: | Mozilla/5.0 (X11; U; Linux i686 (x86_64); en-US; rv:1.8.1.18) Gecko/20081105 Thunderbird/2.0.0.18 Mnenhy/0.7.5.0 |
On 05/25/2009 02:06 PM, Ron Norman wrote:
Other things to consider: A LINKAGE SECTION item can be declared as 01 MYPARAM PIC X ANY LENGTH. And the subroutine code can then do MOVE LENGTH OF MYPARAM TO XXX. I.e. the length of each parameter also needs to be encoded someplace. Also you can do interesting things like SET ADDRESS OF MYPRMX TO ADDRESS OF MYOTHERDATA. CALL "XYZ" USING MYPRMX, LOCALDATA. Also, subroutines may be called with different number of parameters on different calls to the same routine. (This was common practice on S/370 systems.) I have OpenCOBOL as a supported COBOL compiler for my company's OLTP system and being able to pass different number of parameters to our API library (all C code) and having it be able to figure out how many parameters were passed and the length of each parameter is very important. Agreed. The current parsing subsystem is able to convert COBOL to C, so it knows how to translate the parameter passing mechanism. The proposal relies on the current parsing subsystem. Things quickly get complicated... zLINUX is Linux on a Z Series (i.e. S/370) and OpenCOBOL could very easily be running on a zLINUX system, Intel system, Sparc, Itanium, etc... Some people have also ported OpenCOBOL Windows using CYGWIN or something similar. Right. As I mentioned at the close of the last post: this implementation requires an architecture-independent parameter passing mechanism. The current parsing mechanism handles the translation to C for variable length argument lists as well as call by reference. To emphasize the point: the proposal requires the development of an architecture-independent parameter passing mechanism. All of your examples are encompassed in the current parsing subsystem. Each production referenced in the parser would be addressed in the analysis and design tasks. Cheers, jec Cheers Ron Norman -----Original Message----- From: Jeff Chimene [mailto:address@hidden] Sent: May 25, 2009 4:40 PM To: Ron Norman; address@hidden Subject: Re: [open-cobol-list] Proposed addition to OpenCOBOL: an object module generator On 05/25/2009 12:04 PM, Ron Norman wrote:Ok, I think I understand... So if the hardware was say an IBM 370, the code would be somethinglike:ENTRY MAIN MAIN DS 0A DC A(SETUP) DC A(PERFORMA) DC A(PERFORMB) END PERFORMA DS 0A DC A(CALL) DC A(MYRTN) DC A(P1) DC A(P2) DC A(NULL) . . . Plus much much more detail. Then your "inner interpreter" would basically trip thru the code calling routines and passing parameters as required for things like MOVE, COMPUTE, CALL, READ/WRITE etc... Do I have it right?My original example needs amending to address your example. My original example was all calls to cob1, with no concern for user-written code. As such, it was a bad example. Let me try and merge the two. this example mixes code and data, which shouldn't happen in practice. MAIN DS 0A DC A(COBRTL_SETUP) DC A(COBRTL_PUSH) one example DS 8F ... of parameter passing DC A(COBRTL_PUSH) DS 6B DC A(COBRTL_PERFORM) DC A(PERFORMA) P1 DS 8F P2 DS 4B DC A(COBRTL_PERFORM) yet another example of DC A(PERFORMB) ... parameter passing DC A(P1) DC A(P2) ENTRY PERFORMA P1 DS 0F P2 DS 0B DC A(COBRTL_GET_ARG_WORD) These GET_ARG routines either find their DC A(P1) ... values on a virtual stack DC A(COBRTL_GET_ARG_HALFWORD) ... or by referring to the call linkage pointer DC A(P2) DC A(COBRTL_PERFORM) DC A(MYRTN) DC A(P1) DC A(P2) DC A(NULL) COBRTL_PERFORM would read its arguments using an artificial, architecture-independent parameter passing mechanism. Extending the example to include parameters is important. It's also important that your example is the S/370 as it is an architecture that doesn't have a stack. It's clear that this design will have to include a parameter passing mechanism that doesn't rely on a hardware stack. There could be no architecture-specific parameter passing mechanisms. It may be that such a concession is a deal-breaker.------------------------------------------------------------------------*From:* Jeff Chimene [mailto:address@hidden] *Sent:* May 25, 2009 2:47 PM *To:* Ron Norman *Subject:* Re: [open-cobol-list] Proposed addition to OpenCOBOL: an object module generator HI Ron: For future, I'd prefer to keep this discussion on the list. Other comments in line. On 05/25/2009 11:32 AM, Ron Norman wrote: The module format COFF vs ELF is a minor issue compared to whichmachineinstruction set would be used. Agreed. That's why I omitted it from the proposal. In the proposed model, the executable code is all C. As for the proposed object file, when decoded to assembler, it might look something like: .entry MAIN .extern SETUP .extern PERFORMA .extern PERFORMB .end The concept is that a C routine (aka the "inner interpreter") calls the routines (also written in C, in the current cob1 library and a TBDlibrary). There needs to be additional code that replaces the current C code emitted by the COBOL compiler, The new C code will handle branching and data structures (data definition, working storage). Thatnew code would be in the TBD library. To reiterate, all code that must be portable is written in GNU C. The object file output by the COBOL compiler contains no executable code, only pointers to routines in cob1 and another TBD library. The organization of these pointers is commonly referred to as a "thread". Cheers, jec -----Original Message----- From: Jeff Chimene [mailto:address@hidden] Sent: May 25, 2009 11:16 AM To:Ron Norman Subject: Re: [open-cobol-list] Proposed addition to OpenCOBOL: an objectmodule generator On Mon, May 25, 2009 at 6:15 AM,Ron Norman <address@hidden><mailto:address@hidden> wrote:Just my 2 cents, but GNU C does generate object modules on theplatform inuse.Agreed. One of the proposal's goals is to generate object files that represent compiled COBOL code as opposed to compiled C code.If OpenCOBOL generated the object module directly then it needs todeal withthe many different hardware platforms, Intel x86,Itanium,PA-Risc,PowerPC,SPARC, etc...The object module format would be COFF. That /should/ be portable across those systems.GNU C does have some internal features which seem to allow GOTO andcomputedGOTO to be implemented.Agreed. However, I don't suggest transforming the current C code into threaded C code. I am suggesting using the theaded model in object files. The run-time system uses the GNU C syntax for indirect jumps to follow the thread.Roger While would know better than I.I waiting for feedback from him on this proposal. |
[Prev in Thread] | Current Thread | [Next in Thread] |