Re: Compiler in its own language (Was: Re: Play 20K expressions again, Sam)

[prism!jm59@gatech.edu (MILLS,JOHN M.)]

In article <199209100637.AA13126@cnam.cnam.fr> bortz@CNAM.CNAM.FR writes:
>...
  [ lead-up discussion deleted ]

>IMHO, it is better to write the compiler in the language it compiles,
>whatever is this language. 
>It has several technical benefits:
(1) >- any improvment in the optimizer will optimize the compiler as well,
(2) >- this is the greatest test suite you can dream of: if it can compile
    >an Ada compiler, it will compile everything :-)
>and a commercial one:
(3) >- it states clearly that you trust the language you try to sell.

Without having come _close_ to writing a compiler or interpreter, I have
some very serious reservations about this reasoning.

Programming languages have various ranges of application, strengths and
weaknesses, and tradeoffs of power vs. mastery for effective use.  I
couldn't select only one language to do all types of programming and feel
I had always the best tool.  (Naturally, I think the USDoD's attempt to do
this with Ada forces some poor and expensive design decisions, some of which
were on my projects. .. Sorry, no details.)  C was originally developed to
write operating systems, and its relative simplicity, implementation on
processors from 8080s to Crays (and targeting for an even _broader_ range),
led to its use for a lot of programming, development of C++ and a wide range
of programming support tools, etc.

Just because one needs (for example) a better FORTRAN compiler doesn't imply
the compiler should be written in FORTRAN.  Some BASIC interpreters were
written in [compiled] BASIC, but they were generally not the good ones.

For all I know, Ada may _be_ the best language to write a compiler;  in that
case it should be considered for Pascal compilers, etc.  My plaint is with
the assumption that a single language has the qualities for any and all
applications.

I have seen OOP (in C++) used to provide special operations and data types
needed in DSP programming, allowing DSP code to be written in "near C."
Perhaps OOP will evolve some tool sets with [largely] common user and
programmer interfaces for _whatever_ language is best for the rest of the
program.  That would help both camps, and is probably happening with
windows bindings, as we type.

>I tried to list the languages used by some compilers. (If someone wants
>to complete/correct this list?)
>
>Compiler                 Language
>
>Verdix                   C + now partly in Ada
>Telesoft                 Ada
>Alsys                    Ada
>DEC                      Bliss + Ada (unknown proportions)
>Meridian                 ?
>Ada-Ed, first version    SETL
>Ada-Ed, second version   C
>Systeam                  ?
>GNU Ada                  C, but may be an Ada version (?)
>

Several surveys have compared these various compilers, and I don't
recall any correspondence between user satisfaction and compiler source
language.  Many users (including our team) report satisfaction with the
DEC product, in our case partly because of its close correspondence to
the XDAda for our MC680XX targets.  We also use the RTAda version of the
Telesoft product, and found our development process much more painful
using that package.  Depending on your benchmarks, an otherwise excellent
compiler may be unacceptable:  if you _need_ 100 microsec. interrupts and
the package only turns around in 250, you're stuck!

One size _doesn't_ fit all, even in the same programming language.

Regards --jmm--

-- 
John M. Mills, SRE; Georgia Tech/GTRI/TSDL, Atlanta, GA 30332
uucp: ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!jm59
Internet: jm59@prism.gatech.edu
 "f U cn rd dis, U mst uz Unix!!!"  ...jaw