Digital Mars - digitalmars.D.announce - D compiler as part of GCC

Jerry Quinn <jlquinn optonline.net> writes:

[also posted to D.gnu]

Hi, folks,

I'm interested in creating a D front end for GCC that would be part of the GCC
codebase.  My feeling is that a GDC that is part of GCC distributions will
likely have more life than one that must be updated whenever a new GCC release
comes out.  As with linux kernel in-tree drivers being kept up to date, an
integrated GDC would tend to move forward as well.

To do this though, copyright on the code must be assigned to the FSF.  This
means that even though the DMD front end sources are licensed under the GPL,
they cannot be directly used to write this front end as the copyright is owned
by DigitalMars.  Everyone who contributes code must not look at the DMD
compiler source code to avoid accidentally contributing code illegally. 
Therefore, this will be a completely new implementation of D.

The obvious disadvantage of doing this is that it will be a slow process to get
to a working D compiler.  However, one advantage to the D world is firming up
and validating the language specification so that the language is not defined
by what the DMD compiler does.

My personal desire is to implement (and track) the 2.0 language since I would
like to see that feature set available through GCC.  Second, by the time a
working front end becomes part of GCC, the 2.0 language will likely be complete.

One question I have (of many) is whether a different name should be used.  If
this is called GDC there will be some confusion with the current GDC.  What
thoughts do you all have?

In general is there interest in this project, especially contributing to it?

Thanks,
Jerry

Jan 17 2010

Walter Bright <newshound1 digitalmars.com> writes:

Jerry Quinn wrote:
 My personal desire is to implement (and track) the 2.0 language since
 I would like to see that feature set available through GCC.  Second,
 by the time a working front end becomes part of GCC, the 2.0 language
 will likely be complete.
 
 One question I have (of many) is whether a different name should be
 used.  If this is called GDC there will be some confusion with the
 current GDC.  What thoughts do you all have?
 
 In general is there interest in this project, especially contributing
 to it?


Obviously, I can't help you with a clean room reimplementation, but I do 
wish you every success with the project. I can also help with 
specification problems that will inevitably arise.

Jan 17 2010

Eldar Insafutdinov <e.insafutdinov gmail.com> writes:

bearophile Wrote:

 Jerry Quinn:
 I'm interested in creating a D front end for GCC that would be part of the GCC
codebase.


 What about helping LDC devs create a good D2 implementation instead? It's
probably 1/5 or 1/10 of the work you think about, because lot of work is
already done, and surely some people will help you (me too).
 
 There's Dil, DMD, GDC, LDC, D#, etc, but one good, debugged and well
optimizing fully open source D2 compiler is much better than ten broken and/or
badly optimizing D compilers.
 
 Bye,
 bearophile


I agree that having such a good intent the author of the post should better
concentrate his effort on helping GDC/LDC. LDC took couple of years to become
usable, and you have to consider that they took an existing front-end.

Also what I think even when you complete this project, it is not only the
licensing issues that are preventing GDC from being included into GCC. They
will do that only if they are interested in this project, as it requires
maintenance. They will not update GCC-D frontend with every release of GCC just
because it is a part of it.

Having a solid GDC implementation you can be sure that it will be included in
distributions (Debian had GDC for quite a long time).

Jan 18 2010

Eldar Insafutdinov <e.insafutdinov gmail.com> writes:

Nick Sabalausky Wrote:

 "Eldar Insafutdinov" <e.insafutdinov gmail.com> wrote in message 
 news:hj2njd$o1g$1 digitalmars.com...
 Having a solid GDC implementation you can be sure that it will be included 
 in distributions (Debian had GDC for quite a long time).


 "had"? Is that a typo or did they drop it? 
 
 


Sorry for confusion, I meant it had when I checked it last time. I never used
GDC and I believe not many people do, as D2 went to far away since the last
front-end update, and for D1 a lot of people prefer LDC.

Jan 19 2010

Jerry Quinn <jlquinn optonline.net> writes:

Eldar Insafutdinov Wrote:

 bearophile Wrote:
 
 Jerry Quinn:
 I'm interested in creating a D front end for GCC that would be part of the GCC
codebase.


 What about helping LDC devs create a good D2 implementation instead? It's
probably 1/5 or 1/10 of the work you think about, because lot of work is
already done, and surely some people will help you (me too).




One reason is that I'm already positioned to contribute code to GCC. and it is
more difficult for me to become an LDC dev.  ANother is that GCC has very broad
backend support.  I know LLVM backend support is expanding but it still has
some distance to go.  GCC is also the default compiler for many Linux
distributions, and D be part of that may help it propagate.

I also do think that the construction of another front end would provide
positive benefit to the D community by improving the language specification and
separating it from implementation.  Of course that's only true if this succeeds
:-)

There's also a benefit to the GCC project in terms of improving the docs on the
frontend interface.  That's already happened as I've tried to figure out how it
works :-)  But that's not so relevant to the folks here.

 There's Dil, DMD, GDC, LDC, D#, etc, but one good, debugged and well
optimizing fully open source D2 compiler is much better than ten broken and/or
badly optimizing D compilers.
 
 Bye,
 bearophile


 I agree that having such a good intent the author of the post should better
concentrate his effort on helping GDC/LDC. LDC took couple of years to become
usable, and you have to consider that they took an existing front-end.
 
 Also what I think even when you complete this project, it is not only the
licensing issues that are preventing GDC from being included into GCC. They
will do that only if they are interested in this project, as it requires
maintenance. They will not update GCC-D frontend with every release of GCC just
because it is a part of it.


This is very true.  It would require people to be interested in continuing it's
existence.

 Having a solid GDC implementation you can be sure that it will be included in
distributions (Debian had GDC for quite a long time).


In my mind, the endgame for GDC would be to have the work integrated into the
official GCC sources.  That would provide the similar benefits to the ones I'm
chasing.  Everyone who touched GDC would have to assign their code to the FSF
and the DMD sources would also have to be assigned.  Perhaps that's the right
answer in the end but I don't know.  It does seem to be a substantial effort to
make that happen.

Jerry

Jan 18 2010

Steve Teale <steve.teale britseyeview.com> writes:

Walter Bright Wrote:

 Leandro Lucarella wrote:
 I agree that embarking a new front-end will be a huge effort that probably
 will end up abandoned before it's completed, unless there is some economic
 sponsorship or something, but having a front-end which copyright can be
 given to the FSF is a necessary condition to merge GDC (or whatever it's
 named) to GCC.


 Will they take a fork of the dmd source, such that they own the 
 copyright to the fork and Digital Mars still has copyright to the original?


Go for it Walter - the paths to fame are incomprehensible. Also, you'll still
be faster than they are!

Jan 19 2010

Walter Bright <newshound1 digitalmars.com> writes:

Leandro Lucarella wrote:
 Walter Bright, el 23 de enero a las 12:54 me escribiste:
 Jerry Quinn wrote:
 Walter Bright Wrote:
 Will they take a fork of the dmd source, such that they own the
 copyright to the fork and Digital Mars still has copyright to
 the original?



 The answer appears to be yes:

 http://gcc.gnu.org/ml/gcc/2010-01/msg00430.html

 Jerry




 Great news indeed! Since DMD FE is GPL I think it won't be any trouble to
 fold in the new changes back to GDC as they did (and LDC too), so it won't
 be really a *fork*, right?


Well, still I won't be supporting gdc directly. It would mean a team 
that would be willing to take new DMD FE updates and fold them into GDC, 
and then follow whatever gcc's build and release conventions are.


 Walter, please, please, please let us know how this progresses. Thanks!


Who in the FSF is a contact person about this?

Jan 23 2010

Jerry Quinn <jlquinn optonline.net> writes:

Brad Roberts Wrote:

 On 1/23/2010 4:15 PM, Walter Bright wrote:
 Leandro Lucarella wrote:
 Walter Bright, el 23 de enero a las 12:54 me escribiste:
 Jerry Quinn wrote:
 Walter Bright Wrote:
 Will they take a fork of the dmd source, such that they own the
 copyright to the fork and Digital Mars still has copyright to
 the original?



 The answer appears to be yes:

 http://gcc.gnu.org/ml/gcc/2010-01/msg00430.html

 Jerry


 about!


 Great news indeed! Since DMD FE is GPL I think it won't be any trouble to
 fold in the new changes back to GDC as they did (and LDC too), so it
 won't
 be really a *fork*, right?


 Well, still I won't be supporting gdc directly. It would mean a team
 that would be willing to take new DMD FE updates and fold them into GDC,
 and then follow whatever gcc's build and release conventions are.


 I don't think you got the answer you were looking for.  You got an answer to a
 different question.  If you assign the copyright over to the FSF, they then own
 the code.  You'd have a license to use it as you like in return, but you would
 no longer be the owner.


 Additionally, as pointed out in the gcc  thread, contributions coming into the
 gcc tree wouldn't have anything other than the gpl license attached to them and
 that would likely make them problematic to re-distribute from your tree with
the
 dual gpl/artistic license.
 
 In simpler words, this is still far from straightforward.


I think you're slightly incorrect, Brad.  DigitalMars still owns the copyright
to the original source (call it copy A).  A fork (called copy B) is donated to
the FSF.   DigitalMars still gets to make changes to copy A and license them as
it sees fit.  Copy B is part of the GCC codebase and would evolve separately.

Moving changes between them would require the same kind of donation process as
the original transfer.  Folks making changes to the DMD FE would have to
contribute those changes to FSF as well to get them into copy B and vice versa.

My apologies if that's the same as what you said.  I read your comment a couple
of times and was a bit confused.
 
 I'd still love for there to be fewer split efforts on the compiler front, so I
 do encourage trying to find a workable solution.. but tread carefully.


The GCC java front end currently uses the Eclipse compiler to produce bytecode
and then compiles the bytecode to native. But that's the only front end I'm
aware of that isn't fully integrated into the GCC tree.

The D front end doesn't produce a portable intermediate representation like
that so I think it would be harder to always use the latest DMD front end.

I see this possibility as Walter giving GCC a running start so that the D
ecosystem has another viable compiler option available with relatively low
effort.  

In the end, the language spec should be the thing that unifies the D community
rather than the adhoc definition provided by a particular front end
implementation.  It's just a matter of how to get there.

Later,
Jerry

Jan 24 2010

Jesse Phillips <jessekphillips+D gmail.com> writes:

Jerry Quinn wrote:

 I think you're slightly incorrect, Brad.  DigitalMars still owns the copyright
to the original source (call it copy A).  A fork (called copy B) is donated to
the FSF.   DigitalMars still gets to make changes to copy A and license them as
it sees fit.  Copy B is part of the GCC codebase and would evolve separately.

 Moving changes between them would require the same kind of donation process as
the original transfer.  Folks making changes to the DMD FE would have to
contribute those changes to FSF as well to get them into copy B and vice versa.


As best I could tell there were two options, the one Brad was referring
to[1], and the one you asked about.


 In the end, the language spec should be the thing that unifies the D community
rather than the adhoc definition provided by a particular front end
implementation.  It's just a matter of how to get there.


I think Brad was refering to the "donation" process that is required for
propogating changes from DM to GCC and visa versa. Since GCC will be
using the same front end, it would make since that patches should be
applied to both reducing duplicate effort in fixing bugs.

I think that at this time, contributers to the front end would not have
a problem with making these "donations." However in the feature, you
might see more people contributing to GCC and not want to donate it for
GPL/Artistic... And when that happens I don't think Walter would care
that GCC is getting more attention.


1. http://gcc.gnu.org/ml/gcc/2010-01/msg00432.html

Jan 24 2010

Walter Bright <newshound1 digitalmars.com> writes:

Brad Roberts wrote:
 That's the question that Walter asked to have clarified, but that's not the
 question that was asked.  Asking the gcc developers is also a bad idea since
 they're not lawyers.  The only way to really handle this correctly is to have
 lawyers do the question asking of lawyers.  Having lay people (including
myself)
 as intermediaries and interpreters is just wrong.


You're probably right, the only way to do this is to consult a lawyer. 
That's going to be thousands of dollars. And frankly, I've never worked 
with a lawyer who was willing to commit to any particular legal opinion 
anyway.

Jan 24 2010

Brad Roberts <braddr puremagic.com> writes:

On 1/24/2010 2:13 PM, Walter Bright wrote:
 Brad Roberts wrote:
 That's the question that Walter asked to have clarified, but that's
 not the
 question that was asked.  Asking the gcc developers is also a bad idea
 since
 they're not lawyers.  The only way to really handle this correctly is
 to have
 lawyers do the question asking of lawyers.  Having lay people
 (including myself)
 as intermediaries and interpreters is just wrong.


 You're probably right, the only way to do this is to consult a lawyer.
 That's going to be thousands of dollars. And frankly, I've never worked
 with a lawyer who was willing to commit to any particular legal opinion
 anyway.


You're probably versed enough to do the talking for yourself with one of the FSF
lawyers.  Chances are that might actually not cost you anything.

Jan 24 2010