🚀 go-pugleaf

In article <CpJKJz.KBD@news.cis.umn.edu> ianhogg@milli.cs.umn.edu (Ian J Hogg) writes:
>In article <JASON.94May6154541@deneb.cygnus.com> jason@cygnus.com (Jason Merrill) writes:
>>>>>>> Ian J Hogg <ianhogg@milli.cs.umn.edu> writes:
>>
>>> Where does anything say that C and C++ use the same calling convention
>>> for plain functions.  As I've stated before I know of at least one
>>> compiler that passes arguments for C++ linkage functions differently than
>>> C linkage functions.  Are you saying this compiler is broken?
>>
>>I was only referring to my perception of existing practice; evidently I was
>>mistaken.  It does seem strange to me that they would alter the calling
>>convention, but I suppose they had their reasons.

If people followed the language standards, instead of bitching when
they got caught with their hands in the cookie jar, nobody would have
noticed.

(Okay, that's a little strong because most companies do use the same
calling conventions -- probably because they see so many people abusing
it.  However, if you follow the ARM and use "extern C" and never try
to call member functions from C, *you would never notice the "problem"*.)

>It was the Zortech compiler for SYSV/386.  They claimed their "C++" linkage
>parameter passing was more efficient.  

C++ requires all function have prototypes, which allows the compiler
to remove arguments from the stack in the *called* function for any
function with a fixed number of arguments (99%).

Assember code looks like:

   push _var1                foo:  code..
   push _var2                      ret 4
   call foo
   whatever...

C does not require function prototypes (although it will use them if
available) and there is nothing to prevent a developer from specifying
too many or too few parameters.  Therefore, the *calling* function

Assember code looks like:

   push _var2                foo:  code..
   push _var1                      ret
   call foo
   pop 4
   whatever...

The first approach is slightly faster, and is the one normally implemented.
It is also the "pascal" convention available on PC compilers, and this
performance improvement is why the MS/Windows SDK uses the "pascal" 
modifier on all procedures.

BTW, note that the arguments in the second example were pushed on the stack
in reversed order.  In C, since you don't know the number of arguments you
need to push them from right-to-left, so that the left-most argument is
always on the top of the stack.

In contrast, in C++ you (generally) know how many arguments are present
and you can push them in any order.  I believe this also allows a slightly
more efficient "stack frame", but am not sure -- my assembler/HL interfaces
always used the C conventions.

-- 
Bear Giles
bear@fsl.noaa.gov