ARM RDI

ARM RDI

[Mark Mitchell]

In my quest to make things build on Windows, I see that ARM GDB doesn't
quite build for Windows do the fact that rdi-share doesn't build without
angeldll.h.  However, rdi-share does still build on GNU/Linux, where
rdi-share isn't required.

Since RDI has been declared obsolete by ARM, and since the version in
GDB is even more obsolete than that, I'm not sure that there's much
point in continuing to try to support RDI.

There seem to be several options, ordered, IMHO, from best to worst.

1) Remove rdi-share from GDB entirely.

2) Leave it in, but turn it off by default, giving a configure option to
turn it on.

3) Like 2, but have the default be on.

4) Disable it on MinGW hosts, but leave it on for UNIX-like hosts.

5) Fix rdi-share to work on Windows, somehow, without angeldll.h.

6) Nothing.

Thoughts?  Is anyone actively using rdi-share?

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304

Re: ARM RDI

[Richard Earnshaw]

On Mon, 2005-11-28 at 23:15, Mark Mitchell wrote:
> In my quest to make things build on Windows, I see that ARM GDB doesn't
> quite build for Windows do the fact that rdi-share doesn't build without
> angeldll.h.  However, rdi-share does still build on GNU/Linux, where
> rdi-share isn't required.
> 
> Since RDI has been declared obsolete by ARM, and since the version in
> GDB is even more obsolete than that, I'm not sure that there's much
> point in continuing to try to support RDI.
> 
> There seem to be several options, ordered, IMHO, from best to worst.
> 
> 1) Remove rdi-share from GDB entirely.

I think this one is probably best now...  I discussed it with Andrew at
the Summit and we agreed it was the way to go.  But like dejagnu removal
it's taking a long time to finally do it...

R.

Re: ARM RDI

[Daniel Jacobowitz]

On Tue, Nov 29, 2005 at 10:13:34AM +0000, Richard Earnshaw wrote:
> On Mon, 2005-11-28 at 23:15, Mark Mitchell wrote:
> > In my quest to make things build on Windows, I see that ARM GDB doesn't
> > quite build for Windows do the fact that rdi-share doesn't build without
> > angeldll.h.  However, rdi-share does still build on GNU/Linux, where
> > rdi-share isn't required.
> > 
> > Since RDI has been declared obsolete by ARM, and since the version in
> > GDB is even more obsolete than that, I'm not sure that there's much
> > point in continuing to try to support RDI.
> > 
> > There seem to be several options, ordered, IMHO, from best to worst.
> > 
> > 1) Remove rdi-share from GDB entirely.
> 
> I think this one is probably best now...  I discussed it with Andrew at
> the Summit and we agreed it was the way to go.  But like dejagnu removal
> it's taking a long time to finally do it...

Does anyone object?  If not, let's just do it.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: ARM RDI

[Simon Richter]

[-- Attachment #1: Type: text/plain, Size: 741 bytes --]

Hi,

Daniel Jacobowitz wrote:

>>>1) Remove rdi-share from GDB entirely.

>>I think this one is probably best now...  I discussed it with Andrew at
>>the Summit and we agreed it was the way to go.  But like dejagnu removal
>>it's taking a long time to finally do it...

> Does anyone object?  If not, let's just do it.

For a number of CPUs, RDI is the only standard there is; it is the kind 
of "obsolete" where they want people to buy their newer CPUs, not the 
kind where there is a better standard. I'm fairly certain that there are 
several trees where RDI works just fine as there are a number of 
companies selling gdb based solutions for older ARM CPUs; the 
interesting part will be getting those changes back from them.

    Simon

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 307 bytes --]

Re: ARM RDI

[Richard Earnshaw]

On Tue, 2005-11-29 at 14:11, Simon Richter wrote:
> Hi,
> 
> Daniel Jacobowitz wrote:
> 
> >>>1) Remove rdi-share from GDB entirely.
> 
> >>I think this one is probably best now...  I discussed it with Andrew at
> >>the Summit and we agreed it was the way to go.  But like dejagnu removal
> >>it's taking a long time to finally do it...
> 
> > Does anyone object?  If not, let's just do it.
> 
> For a number of CPUs, RDI is the only standard there is; it is the kind 
> of "obsolete" where they want people to buy their newer CPUs, not the 
> kind where there is a better standard. I'm fairly certain that there are 
> several trees where RDI works just fine as there are a number of 
> companies selling gdb based solutions for older ARM CPUs; the 
> interesting part will be getting those changes back from them.

It's also obsolete in the GNU sense of 'nobody has stepped up to
maintain the code'.  


Are you volunteering?

R.

Re: ARM RDI

[Daniel Jacobowitz]

On Tue, Nov 29, 2005 at 03:11:30PM +0100, Simon Richter wrote:
> Hi,
> 
> Daniel Jacobowitz wrote:
> 
> >>>1) Remove rdi-share from GDB entirely.
> 
> >>I think this one is probably best now...  I discussed it with Andrew at
> >>the Summit and we agreed it was the way to go.  But like dejagnu removal
> >>it's taking a long time to finally do it...
> 
> >Does anyone object?  If not, let's just do it.
> 
> For a number of CPUs, RDI is the only standard there is; it is the kind 
> of "obsolete" where they want people to buy their newer CPUs, not the 
> kind where there is a better standard. I'm fairly certain that there are 
> several trees where RDI works just fine as there are a number of 
> companies selling gdb based solutions for older ARM CPUs; the 
> interesting part will be getting those changes back from them.

There are a number of other ways to use RDI besides rdi-share.  I
believe Cygnus produced an RDI<->GDB conversion daemon and it may still
be available from Red Hat; I know CodeSourcery also ships such a
beast:

  http://www.codesourcery.com/gnu_toolchains/arm/faq.html#c_rdi

We're only talking about the in-tree copy of rdi-share, which is a
different beast alltogether.  I don't even think it works with some
current RDI targets.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: ARM RDI

[Mark Mitchell]

Daniel Jacobowitz wrote:

> There are a number of other ways to use RDI besides rdi-share.

Indeed.

Some of the board vendors also ship GDB<->RDI converters.  As you say, I
have very little confidence that the current rdi-share actually works
with most things.

If consensus has been reached to remove rdi-share, may I post a patch
for review?  Or, if we don't think we can get that consensus, how about
at least a patch to make it a configure-time option, off by default?

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304

Re: ARM RDI

[Daniel Jacobowitz]

On Tue, Nov 29, 2005 at 09:54:07AM -0800, Mark Mitchell wrote:
> Daniel Jacobowitz wrote:
> 
> > There are a number of other ways to use RDI besides rdi-share.
> 
> Indeed.
> 
> Some of the board vendors also ship GDB<->RDI converters.  As you say, I
> have very little confidence that the current rdi-share actually works
> with most things.
> 
> If consensus has been reached to remove rdi-share, may I post a patch
> for review?  Or, if we don't think we can get that consensus, how about
> at least a patch to make it a configure-time option, off by default?

Consensus can't be reached in one day :-)

Let's come back to this next week, please.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: ARM RDI

[Mark Mitchell]

Daniel Jacobowitz wrote:

> Consensus can't be reached in one day :-)

Fair enough!

> Let's come back to this next week, please.

Roger that.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304

Re: ARM RDI

[Daniel Jacobowitz]

On Tue, Nov 29, 2005 at 01:03:26PM -0500, Daniel Jacobowitz wrote:
> On Tue, Nov 29, 2005 at 09:54:07AM -0800, Mark Mitchell wrote:
> > Daniel Jacobowitz wrote:
> > 
> > > There are a number of other ways to use RDI besides rdi-share.
> > 
> > Indeed.
> > 
> > Some of the board vendors also ship GDB<->RDI converters.  As you say, I
> > have very little confidence that the current rdi-share actually works
> > with most things.
> > 
> > If consensus has been reached to remove rdi-share, may I post a patch
> > for review?  Or, if we don't think we can get that consensus, how about
> > at least a patch to make it a configure-time option, off by default?
> 
> Consensus can't be reached in one day :-)
> 
> Let's come back to this next week, please.

Only Simon objected, and I haven't seen a reply to either Richard's or
my responses to him.  Richard (the ARM port maintainer) and Andrew
(second-hand via Richard) were in favor.  Anyone else have comments? 
Otherwise, next week I intend to remove the rdi-share subdirectory and
its supporting code.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: ARM RDI

[Simon Richter]

[-- Attachment #1: Type: text/plain, Size: 366 bytes --]

Hi,

Daniel Jacobowitz wrote:

> Only Simon objected, and I haven't seen a reply to either Richard's or
> my responses to him.

I was fully content with the answer, as my objection was based on the 
belief that functional code was to be removed because RDI was declared 
obsolete, not because rdi-share was broken and unused.

So removal is fine with me.

    Simon

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 307 bytes --]

Re: ARM RDI

[Stan Shebs]

Daniel Jacobowitz wrote:

>Otherwise, next week I intend to remove the rdi-share subdirectory and
>its supporting code.
>
Yes, a good idea. Sic transit and all that... It has some small
value as an object lesson in the downsides of importing other
people's libraries, but the changelogs include a laconic account
that is still educational, ahem.

Stan

Re: ARM RDI

[Jim Blandy]

On 11/28/05, Mark Mitchell <mark@codesourcery.com> wrote:
> In my quest to make things build on Windows, I see that ARM GDB doesn't
> quite build for Windows do the fact that rdi-share doesn't build without
> angeldll.h.  However, rdi-share does still build on GNU/Linux, where
> rdi-share isn't required.

You meant "where angeldll.h isn't required", right?

Setting aside questions about whether we care about rdi-share at all
for the moment, the GNU Way to deal with this would be for the
configure script to recognize that angeldll.h isn't available, and
decline to build rdi-share.  When you say "Windows", you mean "Windows
without Cygwin", so I don't know how the configury works there.  Is
there any way to get this behavior?

Re: ARM RDI

[Mark Mitchell]

Jim Blandy wrote:
> On 11/28/05, Mark Mitchell <mark@codesourcery.com> wrote:
> 
>>In my quest to make things build on Windows, I see that ARM GDB doesn't
>>quite build for Windows do the fact that rdi-share doesn't build without
>>angeldll.h.  However, rdi-share does still build on GNU/Linux, where
>>rdi-share isn't required.
> 
> You meant "where angeldll.h isn't required", right?

Yes; sorry.

> decline to build rdi-share.  When you say "Windows", you mean "Windows
> without Cygwin", so I don't know how the configury works there.  Is
> there any way to get this behavior?

Yes, checks for headers work fine, and we know are host OS, there are at
least two ways to do as you suggest, so I should have added another option:

*) Build rdi-share only on non-MinGW hosts.

If we want to keep rdi-share, then that might be a good option.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304