www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - DMD in D?

reply "Lionello Lunesu" <lio lunesu.removethis.com> writes:
Are there any plans or is there already a project for writing DMD in D?

Nothing proves the stability of a compiler more than it being able to 
compile itself. And wouldn't it speed up the creation and maintenance of the 
compiler too?

..But then I found a message from Walter (18 jan '05):

 It wouldn't be hard to translate the DMD front end into D, but then it 
 would
 be very hard to connect it to back ends like gcc's.


So that's the reason D's front-end (../dmd/src) is in C++: because it's 
using the C compiler's back end and that one's in C/C++?

And it's that interfacing D with C++ problem again. If DMD would decorate 
class members the same as DMC (using the same calling convention, etc.), 
would it then be possible to link a .d in a C/C++ project, like DMC?

Lionello.
Mar 17 2005
next sibling parent reply =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Lionello Lunesu wrote:


 So that's the reason D's front-end (../dmd/src) is in C++: because it's 
 using the C compiler's back end and that one's in C/C++?


If DMD was written in D, what would you use to bootstrap it with ?

And you would have all the woes of the current compiler, like
not being able to create shared libraries, missing debug info,
plain old bugs, and other things that make D yet unreleased...

No, I for one think it's a good thing that DMD front-end is in C++.

--anders
Mar 17 2005
next sibling parent reply "Nick Sabalausky" <z a.a> writes:
"Anders F Björklund" <afb algonet.se> wrote in message 
news:d1c3n4$1164$1 digitaldaemon.com...

 Lionello Lunesu wrote:


 So that's the reason D's front-end (../dmd/src) is in C++: because it's 
 using the C compiler's back end and that one's in C/C++?


 If DMD was written in D, what would you use to bootstrap it with ?


The first 'DMD written in D' would be compiled with the latest C/C++-written 
version. Subsequent versions would be compiled with the previous build.

Unless I misunderstand what you mean by 'bootstrap'...?


 And you would have all the woes of the current compiler, like
 not being able to create shared libraries, missing debug info,
 plain old bugs, and other things that make D yet unreleased...

 No, I for one think it's a good thing that DMD front-end is in C++.

 --anders
Mar 17 2005
next sibling parent reply =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Nick Sabalausky wrote:


If DMD was written in D, what would you use to bootstrap it with ?


 
 The first 'DMD written in D' would be compiled with the latest C/C++-written 
 version. Subsequent versions would be compiled with the previous build.


So would that mean that you had to keep *two* D front-ends around?


Just seeemed a bit strange to rewrite the compiler in a different 
language when it was just barely starting to work written in C++... ;-)

Especially when it would be more useful to update it to, for instance,
work with GCC 4.0; instead of just doing a work-a-like but written in D?


Probably could make for a fun experiment or something. Just thought
that there was more urgent things to do before that... Like "D 1.0"?

Or any other of http://www.prowiki.org/wiki4d/wiki.cgi?HelpDProgress

--anders
Mar 17 2005
parent reply Ilya Minkov <minkov cs.tum.edu> writes:
Anders F Björklund wrote:

 So would that mean that you had to keep *two* D front-ends around?


The one in C++ can be abandoned, if we agree to write the D front-end in
a subset of the language that can be compiled with both the C++ and the
new D front-end.

For the one, i would suggest not to rewrite anything into something
hastingly. Walter has quite a particular D-like style, so a tool could
be written to translate the compiler source into D, taking its
particularities in account. Regularly test it if it works, develop it in
parallel to the compiler, and when it's ready, just make a switch. Of
course this is only a very hypothetical suggestion because i bet noone
would volunteer. I know i won't.


 Just seeemed a bit strange to rewrite the compiler in a different 
 language when it was just barely starting to work written in C++... 
 ;-)
 
 Especially when it would be more useful to update it to, for 
 instance, work with GCC 4.0; instead of just doing a work-a-like but 
 written in D?


Why should this be a problem? GCC backend, being written in C, can be
talked to from any language one can compile, when a compatible C
compiler exists. In fact, GNU Pascal frontend is written in Pascal, etc.

I think i recall Walter saying he wants to have D frontend written in D 
someday. However, it's definately not time yet now so i think we can 
pick up on this discussion when we are past the currently awaited 
milestones.

-eye
Mar 17 2005
parent =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Ilya Minkov wrote:


 Especially when it would be more useful to update it to, for instance, 
 work with GCC 4.0; instead of just doing a work-a-like but written in D?


 
 Why should this be a problem? GCC backend, being written in C, can be
 talked to from any language one can compile, when a compatible C
 compiler exists. In fact, GNU Pascal frontend is written in Pascal, etc.


It's not really a problem, like you say, just thought it be more work?

Had the interface been in C++ instead, matters would be different...
But interfacing with regular C, that D can do without too much hassle.


 I think i recall Walter saying he wants to have D frontend written in D
 someday. However, it's definately not time yet now so i think we can
 pick up on this discussion when we are past the currently awaited
 milestones.


Agreed. I'm not trying to stop it at all, as long as GDC still works :-)

--anders
Mar 17 2005
prev sibling parent Charles Hixson <charleshixsn earthlink.net> writes:
Nick Sabalausky wrote:

 "Anders F Björklund" <afb algonet.se> wrote in message 
 news:d1c3n4$1164$1 digitaldaemon.com...
 

Lionello Lunesu wrote:


...
 
 The first 'DMD written in D' would be compiled with the latest C/C++-written 
 version. Subsequent versions would be compiled with the previous build.
 
 Unless I misunderstand what you mean by 'bootstrap'...?
 
 

And you would have all the woes of the current compiler, like
not being able to create shared libraries, missing debug info,
plain old bugs, and other things that make D yet unreleased...

No, I for one think it's a good thing that DMD front-end is in C++.

--anders 


 
 
 


That's essentially the approach that Algol took to distribution 
back in the 1960's.  It worked, too.  (Algol had a minimal subset 
compiler written in assembler.  The minimal subset was used to 
write a compiler for the full language.  The full language was 
used to create the libraries...)

Algol was the first time I heard about the bootstrap process.  It 
was quite successful yielding such descendants as Pascal and Ada, 
and, less completely, any language that uses block structured code.
Mar 17 2005
prev sibling next sibling parent reply "Martin M. Pedersen" <martin moeller-pedersen.dk> writes:
"Anders F Björklund" <afb algonet.se> skrev i en meddelelse 
news:d1c3n4$1164$1 digitaldaemon.com...

 Lionello Lunesu wrote:

 If DMD was written in D, what would you use to bootstrap it with ?

 And you would have all the woes of the current compiler, like
 not being able to create shared libraries, missing debug info,
 plain old bugs, and other things that make D yet unreleased...


I don't see any of these as show-stoppers. What you need is one stable 
implementation on one platform, then you can quite easily cross-compile if 
you accept to generate eg. C as intermediate code. It is my impression that 
the DMD implementation has become quite stable, and is up for the task. I 
don't see how shared libraries fit in, the level of debug information is 
adequate (on Windows at least (which is good enough)), and trouble with 
immature libraries is not really a problem for a compiler, as its 
dependencies on libraries are quite small.


 No, I for one think it's a good thing that DMD front-end is in C++.


It has been a natural place to start, but a move to D would IMHO be a 
natural evolotion, and a good proof of concept. Walter may have his reasons 
to keep the implementation in C++ for now. I can only guess on what they 
could be, but guesses would include:

    - other issues are deemed to have higher priority.
    - the interface to his backend isn't trivial.
    - he needs to make a living, and breaking the dependency on the backend 
is not good for business
       (and definitely not a priority).

Anyway, who says only Walter can do it? As I remember, he has encouraged a 
project like this.

Regards,
Martin
Mar 17 2005
next sibling parent =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Martin M. Pedersen wrote:


 I don't see any of these as show-stoppers. What you need is one stable 
 implementation on one platform, then you can quite easily cross-compile if 
 you accept to generate eg. C as intermediate code. It is my impression that 
 the DMD implementation has become quite stable, and is up for the task. I 
 don't see how shared libraries fit in, the level of debug information is 
 adequate (on Windows at least (which is good enough)), and trouble with 
 immature libraries is not really a problem for a compiler, as its 
 dependencies on libraries are quite small.


True, just thought that it was more of a "D 2.0" thing to be doing ?

Like at least finalize the language specification before starting
a rewrite of the compiler for it in itself ? Maybe I'm a sceptic.


No, I for one think it's a good thing that DMD front-end is in C++.


 
 It has been a natural place to start, but a move to D would IMHO be a 
 natural evolotion, and a good proof of concept. Walter may have his reasons 
 to keep the implementation in C++ for now. 


I'm glad that he did, and that he then decided to make it Open Source
so that the frontend could be ported to GCC backend and made portable.

If not, I wouldn't be using D myself, so for that I am thankful...

--anders
Mar 17 2005
prev sibling parent "Walter" <newshound digitalmars.com> writes:
"Martin M. Pedersen" <martin moeller-pedersen.dk> wrote in message
news:d1cpb7$1pg9$1 digitaldaemon.com...

 It has been a natural place to start, but a move to D would IMHO be a
 natural evolotion, and a good proof of concept. Walter may have his


reasons

 to keep the implementation in C++ for now. I can only guess on what they
 could be, but guesses would include:

     - other issues are deemed to have higher priority.


Yup.


     - the interface to his backend isn't trivial.


Right. There are a number of .h files, the back end is in C++, and it would
be a fair chunk of work to interface it. I've got far too many other things
to do right now.
Mar 17 2005
prev sibling parent "Zz" <Zz Zz.com> writes:
"Anders F Björklund" <afb algonet.se> wrote in message 
news:d1c3n4$1164$1 digitaldaemon.com...

 Lionello Lunesu wrote:


 So that's the reason D's front-end (../dmd/src) is in C++: because it's 
 using the C compiler's back end and that one's in C/C++?


 If DMD was written in D, what would you use to bootstrap it with ?

 And you would have all the woes of the current compiler, like
 not being able to create shared libraries, missing debug info,
 plain old bugs, and other things that make D yet unreleased...

 No, I for one think it's a good thing that DMD front-end is in C++.


How about writing a small compiler in something like c that only covers a 
small part of the D and use that compiler to create the full version.

I know of a modula-2 compiler that is built this way.

Zz
Mar 17 2005
prev sibling next sibling parent reply Stewart Gordon <smjg_1998 yahoo.com> writes:
Lionello Lunesu wrote:

 Are there any plans or is there already a project for writing DMD 
 in D?
 
 Nothing proves the stability of a compiler more than it being able 
 to compile itself.  And wouldn't it speed up the creation and 
 maintenance of the compiler too?
 
 ..But then I found a message from Walter (18 jan '05):
 

 It wouldn't be hard to translate the DMD front end into D, but 
 then it would be very hard to connect it to back ends like gcc's.


 
 So that's the reason D's front-end (../dmd/src) is in C++: because 
 it's using the C compiler's back end and that one's in C/C++?


I'm guessing that for the time being, any project to write a D compiler 
in D would remain a separate project from DMD itself.  It could be 
designed to be linked with a D back-end, which would open up the 
potential for compilers to be written in D.

The obvious way to bootstrap D on a new platform would be to keep hold 
of the C++-written DMD until a better solution is found.  FTM, what does 
a new platform's first C compiler tend to be written in?


 And it's that interfacing D with C++ problem again.  If DMD would 
 decorate class members the same as DMC (using the same calling 
 convention, etc.), would it then be possible to link a .d in a 
 C/C++ project, like DMC?


http://www.digitalmars.com/d/overview.html

"Features to Drop"


just too complicated - properly supporting it would essentially imply
making D a full C++ compiler too."

Stewart.

-- 
My e-mail is valid but not my primary mailbox.  Please keep replies on 
the 'group where everyone may benefit.
Mar 17 2005
parent reply =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Stewart Gordon wrote:


 The obvious way to bootstrap D on a new platform would be to keep hold 
 of the C++-written DMD until a better solution is found.  FTM, what does 
 a new platform's first C compiler tend to be written in?


Usually bootstrapped using assembler. Or cross-compiled, if available.

Side note:
I thought that D was written in plain old C first, because of the 
misleading .c suffix that Walter is using for the C++ source files...

--anders
Mar 17 2005
parent Mark T <Mark_member pathlink.com> writes:
In article <d1c9o0$1670$2 digitaldaemon.com>,
=?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= says...

Stewart Gordon wrote:


 The obvious way to bootstrap D on a new platform would be to keep hold 
 of the C++-written DMD until a better solution is found.  FTM, what does 
 a new platform's first C compiler tend to be written in?


Usually bootstrapped using assembler. Or cross-compiled, if available.


I would say that a C cross compiler would be the first choice for a new CPU with
a smattering of assembler.
Mar 20 2005
prev sibling parent "Dave" <Dave_member pathlink.com> writes:
"Lionello Lunesu" <lio lunesu.removethis.com> wrote in message 
news:d1c2f3$109f$1 digitaldaemon.com...

 Are there any plans or is there already a project for writing DMD in D?

 Nothing proves the stability of a compiler more than it being able to 
 compile itself. And wouldn't it speed up the creation and maintenance of 
 the compiler too?

 ..But then I found a message from Walter (18 jan '05):

 It wouldn't be hard to translate the DMD front end into D, but then it 
 would
 be very hard to connect it to back ends like gcc's.


 So that's the reason D's front-end (../dmd/src) is in C++: because it's 
 using the C compiler's back end and that one's in C/C++?

 And it's that interfacing D with C++ problem again. If DMD would decorate 
 class members the same as DMC (using the same calling convention, etc.), 
 would it then be possible to link a .d in a C/C++ project, like DMC?

 Lionello.



Historically, I think "version 2" or equivalent of various language 
compilers have been ported to and bootstrapped with the latest v1.x of the 
language and reference compiler because at that time both the language and 
compiler would be pretty stable.

Maybe what Walter has planned is to port both front-end and back-end of the 
compiler for v2.0? And that would open up trying more aggressive 
optimizations specialized for D w/o changing the C/++ backend which is 
specialized for those languages.

A branch strategy like that would make a lot of sense, IMHO.

As for the "..very hard to connect to backends like GCC's", I think what was 
being referred to there was that building a frontend for GCC relies on so 
many C headers, GCC specific build options, etc.. that doing the frontend in 
another language that isn't in the base bootstrap configuration would be 
pretty tough if not impossible. Again, a stable v1.x of the GDC compiler 
included in the base GCC distro. would make that much more straightforward. 
Heck, AFAIK the GCJ frontend is still written completely in C and that's 
been around for years.

- Dave
Mar 17 2005