[COFF] [TUHS] Unix quix

[COFF] [TUHS] Unix quix

[peter]

=> coff since it's non-Unix

On 2020-Jan-22 13:42:44 -0500, Noel Chiappa <jnc at mercury.lcs.mit.edu> wrote:
>Pretty interesting machine, if you study its instruction set, BTW; with no
>stack, subroutines are 'interesting'.

"no stack" was fairly standard amongst early computers.  Note the the IBM
S/360 doesn't have a stack..

The usual approach to subroutines was to use some boilerplate as part of the
"call" or function prologue that stashed a return address in a known
location (storing it in the word before the function entry or patching the
"return" branch were common aproaches).  Of course this made recursion
"hard" (re-entrancy typically wasn't an issue) and Fortran and Cobol (at
least of that vintage) normally don't support recursion for that reason.

-- 
Peter Jeremy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 963 bytes
Desc: not available
URL: <http://minnie.tuhs.org/pipermail/coff/attachments/20200123/49fbaae1/attachment.sig>

[COFF] [TUHS] Unix quix

[clemc]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 3888 bytes --]

FWIW: System 360 uses what was called the 'push down save area' as part of
the function calling convention.  Each routine declared and creates local
storage for the registers.   Being reentrant was certainly possible and
many languages such as the Algol and APL families were.   It's funny since
S/360 was the first architecture I knew deeply (*i.e.* got paid to
program), and working on support for York/APL at the time (and began
learning the Univac 1100 too), I just thought this was natural until I
began to learn about other processors ISA's like the PDP-11 family that had
an SP.

I remember, thinking -- this is so cool.

But as you said, originally early Fortran and Cobol didn't require same.
The typical calling conventions was something like this was pretty standard
for the S/360:

WORKAREA  DSECT ,                    Reentrant work area (stack like function)
          DS    18F                  Save area
FIELD1    DS    F                    Some variable
FIELD2    DS    F                    Another variable
WORKLEN   EQU   *-WORKAREA           Length of reentrant work area

SUBRTN1   RSECT ,                    HLASM will perform reentrant checking
          STM   R14,R12,12(R13)      Save registers at entry
          LR    R12,R15              Set code base register
          USING SUBRTN1,R12          Establish code addressability
          LGHI  R0,WORKLEN           Get length of reentrant work area
          STORAGE OBTAIN,            Obtain reentrant work area
          X
                LENGTH=(0)           ..Length is in R0
          ST    R1,8(,R13)           Forward chain in prev save area
          ST    R13,4(,R1)           Backward chain in next save area
          L     R14,20(,R13)         Get R1 at entry (parameters)
          LR    R13,R1               Set up new save area/reentrant workarea
          USING WORKAREA,R13         Establish work area addressability
          LM    R2,R3,0(R14)         Get addresses of parameters
          STM   R2,R3,FIELD1         Save parameter addresses for later
…
***    Subroutine Logic goes here
…
          LR    R1,R13               Address to be released
          L     R13,4(,R13)          Address of prior save area
          LGHI  R0,WORKLEN           Length of storage to release
          STORAGE RELEASE,           Release reentrant work area
          X
                ADDRESS=(1),         ..Address in R1
          X
                LENGTH=(0)           ..Length in R0
          LM    R14,R12,12(R13)      Restore registers
          OI    15(R13),X'01'        This bit on means this save area
is inactive
          BR    R14                  Return to caller



On Thu, Jan 23, 2020 at 12:47 AM Peter Jeremy <peter at rulingia.com> wrote:

> => coff since it's non-Unix
>
> On 2020-Jan-22 13:42:44 -0500, Noel Chiappa <jnc at mercury.lcs.mit.edu>
> wrote:
> >Pretty interesting machine, if you study its instruction set, BTW; with no
> >stack, subroutines are 'interesting'.
>
> "no stack" was fairly standard amongst early computers.  Note the the IBM
> S/360 doesn't have a stack..
>
> The usual approach to subroutines was to use some boilerplate as part of
> the
> "call" or function prologue that stashed a return address in a known
> location (storing it in the word before the function entry or patching the
> "return" branch were common aproaches).  Of course this made recursion
> "hard" (re-entrancy typically wasn't an issue) and Fortran and Cobol (at
> least of that vintage) normally don't support recursion for that reason.
>
> --
> Peter Jeremy
> _______________________________________________
> COFF mailing list
> COFF at minnie.tuhs.org
> https://minnie.tuhs.org/cgi-bin/mailman/listinfo/coff
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/coff/attachments/20200123/24bc73a9/attachment.html>

[COFF] [TUHS] Unix quix

[clemc]

BTW:  Not to bring up a new rat hole/new argument, but rather remind people
of one of the arguments of the day.  Because of SP or no SP style of
calling, in those days the question came out language design of how the
stack or register call area was maintained:  Was it/should it be the
responsibility of the calling routine or the called subroutine.  Things
like Fortran's entry statement and other non-local goto's,
co-routines/setjmp/longjmp added fuel to the fire in the argument.

As a language designer/implementor at the time, someone like Steve Johnson
and Doug McIlroy were undoubtedly mixed up in the argument and had
opinions.  I was certainly too inexperienced then to understand what the
fight was about.   But I do remember the fight and how the issues were
taught/discussed at CMU when I was a student.  I admit I was quickly sucked
into C's way of handling things, as it seemed simple/clear/understandable
to me, but not being a language person I probably did not (maybe still do
not) value some the arguments for the other side.

Clem
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/coff/attachments/20200123/2f617b8b/attachment.html>

[COFF] [TUHS] Unix quix

[doug]

Clem wrote, " the question came out language design of how the
stack or register call area was maintained:  Was it/should it be the
responsibility of the calling routine or the called subroutine.  Things
like Fortran's entry statement and other non-local goto's,
co-routines/setjmp/longjmp added fuel to the fire in the argument."

I don't remember much prospective discussion of the subject, but
lots of retrospective discussion of techniques that didn't
quite work, going right back to Lisp's A-list and the ensuing
"funarg problem". The first PL/I out of Hursley got it wrong, too,
but they were able to correct it before the bad way became
ingrained. In C, which had no need for closures (thunks),
efficiency was the main concern in choice of stack protocol.
The only delicate situation, longjump, was sidestepped by
doing it via library rather than language.

Doug

[COFF] [TUHS] Unix quix

[krewat]

On 1/23/2020 12:27 AM, Peter Jeremy wrote:
> "no stack" was fairly standard amongst early computers.  Note the the IBM
> S/360 doesn't have a stack..
You can make a stack out of anything ;)

[COFF] [TUHS] Unix quix

[bakul]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 2569 bytes --]

I read that thunks came about as an implementation of call by name
for Algol. They were basically argument-less closures. Where as a
general closure may accept args as well as refer to values in its
environment (including those in its lexically enclosing function). I
believe at one point gcc used (argument-less) thunks when it added
support for nested functions so as to continue using a single ptr as a
function argument. A proper closure representation needs a pair (a
fun ptr & a ptr to env). The code for all this was stored on the stack &
the thunk address actually pointed to this stacked code. I guess that
had to be abandoned when storing code on stack was found to be a huge
security hole (all because of C’s lax type system & stacks growing down
to smaller addresses).

Nicklaus Wirth carefully limited the scope of functional arguments in Pascal
so as to avoid the “funarg problem” and GC. I think at least by 1975 (when I
first studied software) it was clear how to implement “reentrant” functions;
It is just that not everyone was convinced of the value of it. Even modern
RISC processors don’t actually provide any support for stacks and it is just
their “calling convention” that designates a specific GPR to be a stack ptr!

No one questions the value of subroutines (even though the register to
memory access ratio is soooo much worse now!) but closures and GC are
still not so well accepted.

On Jan 24, 2020, at 2:07 PM, Doug McIlroy <doug at cs.dartmouth.edu> wrote:
> 
> Clem wrote, " the question came out language design of how the
> stack or register call area was maintained:  Was it/should it be the
> responsibility of the calling routine or the called subroutine.  Things
> like Fortran's entry statement and other non-local goto's,
> co-routines/setjmp/longjmp added fuel to the fire in the argument."
> 
> I don't remember much prospective discussion of the subject, but
> lots of retrospective discussion of techniques that didn't
> quite work, going right back to Lisp's A-list and the ensuing
> "funarg problem". The first PL/I out of Hursley got it wrong, too,
> but they were able to correct it before the bad way became
> ingrained. In C, which had no need for closures (thunks),
> efficiency was the main concern in choice of stack protocol.
> The only delicate situation, longjump, was sidestepped by
> doing it via library rather than language.
> 
> Doug
> _______________________________________________
> COFF mailing list
> COFF at minnie.tuhs.org
> https://minnie.tuhs.org/cgi-bin/mailman/listinfo/coff