gforth missing libffcall1 dependancy

[Jason Woofenden]

Binary package hint: gforth

gforth 0.6.2-7.2 is built without ffcall support. This makes it nearly impossible to use libraries.

Solution: install libffcall1-dev before compiling gforth

[Luca Falavigna (dktrkranz)]

Which issues did you encounter without ffcall support? Could this be forwarded to Debian too?

[Jason Woofenden (jason210)]

Without ffcall you cannot use libraries, so eg it's almost impossible to write graphical programs or do networking.

[James Westby (james-w)]

Hi,

Please forward this patch to Debian, as I think having their feedback on it
is important here. I certainly don't know enough to know that this is correct,
so I would want to see this before sponsoring.

Thanks,

James

[Jason Woofenden (jason210)]

Done!

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=500640

[James Westby (james-w)]

Hi,

I just noticed that the Debian maintainer is inactive, so waiting
for a response isn't a good idea.

I had a closer look at the package, and saw that ffcal support was
only recently added, which explains why it's not a dependency, so
I would be happy to add it.

However, my concern is not two-fold, firstly compatibility, secondly
stability. Will switching to ffcal break compatibility with anything? Will
it require a transition? As it is quite new, is is stable, or will it be
riddled with bugs?

I'm subscribing motu-release, as I would like their input on this decision.

Thanks,

James

[Jason Woofenden (jason210)]

I'm pretty sure the additional functionality provided by ffcall in gforth is only activated when specifically requested by the programmer (as it is in some example scripts that come with gforth). So I don't think inclusion of ffcall could possibly break any existing code written for gforth.

If your compatibility concern is with building and/or linking the ffcall library on various architectures, I can't be of much assurance, except to say that it worked flawlessly on x86 and ppc.

Thank you, - Jason

[James Westby (james-w)]

On Tue, 2008-09-30 at 14:52 +0000, Jason Woofenden wrote:
> I'm pretty sure the additional functionality provided by ffcall in
> gforth is only activated when specifically requested by the programmer
> (as it is in some example scripts that come with gforth). So I don't
> think inclusion of ffcall could possibly break any existing code written
> for gforth.

Ok, I think I'm at a point where I would be willing to upload this
after testing. Could you point out a couple of examples from the gforth
source package that would be good tests and how to run them once I have
the package installed? I'd like a couple that use ffcall and a couple
that don't.

I'll give it a day or so for motu-release to give any other opinions.
One member stated on IRC that they considered this to be a change that
didn't need an FFe, so it would just be if they had any concerns over
the change.

Thanks,

James

[Scott Kitterman (kitterman)]

James,

As long as you'll mind after any problems that crop up, I say go ahead.

[Jason Woofenden (jason210)]

I'm attaching two scripts.

Just simple "do math and print result" type things.

One uses ffcall to access libm's sin() and sinf(). The other uses builtin math primitives.

[Jason Woofenden (jason210)]

And here's the one that doesn't need ffcall.

[James Westby (james-w)]

On Tue, 2008-09-30 at 21:58 +0000, Jason Woofenden wrote:
> I'm attaching two scripts.
>
> Just simple "do math and print result" type things.
>
> One uses ffcall to access libm's sin() and sinf(). The other uses
> builtin math primitives.

Great, thanks. I'll take a look at this tomorrow.

Thanks,

James

[James Westby (james-w)]

Hi,

I'm going to have trouble sponsoring this as both the patched gforth
and the one in the archive fail to build on my machine with a Bus Error.
They build fine in my PPA though, so it's something specific to my
machine.

Is someone else willing to look at this?

Thanks,

James

[Cesare Tirabassi (norsetto)]

Nope, it doesn't build. Here attached is the build log if anybody wants to work at it (no clear error message is spit out).

[James Westby (james-w)]

Hi Cesare,

I just pushed the patched version to my PPA and it only failed on amd64,
is that the arch that you were building on?

The failure was apparently the same as in your log.

Thanks,

James

[James Westby (james-w)]

On Wed, 2008-10-01 at 15:55 +0000, James Westby wrote:
> Hi Cesare,
>
> I just pushed the patched version to my PPA and it only failed on amd64,
> is that the arch that you were building on?

Cesare has confirmed that he is on amd64. This would indicate
that the ffcall code is not 64-bit clean. I'm not willing to
upload this change in that case.

Jason, if you would like this uploaded we're going to need a
patch that makes the code 64-bit clean, preferably working
with the developers to check the whole code-base is clean.

Thanks,

James

[James Westby (james-w)]

I'm unsubscribing the sponsors team for now, please re-subscribe
them when ready. I'll stay subscribed in case you have any questions.

Thanks,

James

[Jason Woofenden (jason210)]

Thanks for testing everybody!

I had a conversation last year with the FreeBSD maintainer of gforth, and he couldn't get gforth to build on the build farm with or without libffcall1, though it always built on his home machine (and mine). It could well be that this is a 64-bit issue, but I would like further testing to be sure that it is libffcall that's incompatible, not gforth itself.

Could someone test building gforth both with and without libffcall on an amd64? (gforth automatically builds with libffcall support if it's installed, so you can just run "./debian/rules make" with and without libffcall1-dev installed.)

Or... is there some way I can get access to an amd64 box?

Thanks, -- Jason

[Cesare Tirabassi (norsetto)]

0.6.2-7.3 _without_ ffcall support builds fine on my amd64.

[Cesare Tirabassi (norsetto)]

I have unsubscribed motu-release, not for us anymore.

[David Futcher (bobbo)]

This was fixed in Debian in package version 0.7.0-0.1 and has been included since Karmic. I will therefore mark this bug as Fix Released. If this is still a bug for anyone, please re-open the bug. Thankyou!