Fwd: gitit status update and why are deps needed for binaries

---------- Forwarded message ----------
From: Nicola Squartini
Hi all,
For anybody wondering about the status of gitit: it was removed from [haskell-happstack] on April 7th because of dependency issues (feed
=0.3.6 && <0.4, pandoc >=1.12.4 && <1.14); feed itself was removed because depending on utf8-string < 1.0 which was updated in [haskell-core] to 1-78 [1].
Pandoc is being watched carefully [2][3], but feed doesn't seem really likely to be updated: its Hackage entry [4] points to a GitHub repo that hasn't been updated in 18 months [5], and an issue asking support for utf8-string 1.x is opened there since January 23 [6]. The related discussion makes it clear that feed is leaning towards abandonware.
There is an issue open to have haskell-trustees intervene [7], and a proposition to add it to Stackage [8]. So those are the one to watch concerning gitit's fate.
A question related to the experience with Haskell on Arch: why do pandoc and gitit *binaries* depends on so many haskell packages being installed? It seems that Ubuntu [9] and Debian [10] do without installing the whole Haskell platform and build dependencies (confirmed in a Xubuntu 14.04 VM: only a few libraries are needed). GHC et al. take up so much space (> 1GB) on my server I asked confirmation this was normal on #haskell IRC!
What is the difference in Arch Haskell packaging that causes the need to install all the build depends to solely get the binaries?
Thanks Bastien
[1]
https://github.com/tensor5/haskell-happstack/commit/12356dafcca432ba33fbaa6a... [2] https://github.com/archhaskell/habs/issues/179 [3] https://mail.haskell.org/pipermail/arch-haskell/2015-April/004819.html [4] https://hackage.haskell.org/package/feed [5] https://github.com/sof/ [6] https://github.com/sof/feed/issues/9 [7] https://github.com/haskell-infra/hackage-trustees/issues/12 [8] https://github.com/sof/feed/issues/9#issuecomment-89334563 [9] http://packages.ubuntu.com/vivid/gitit [10] https://packages.debian.org/jessie/gitit _______________________________________________ arch-haskell mailing list arch-haskell@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/arch-haskell

On 14 April 2015 at 02:21, Nicola Squartini
---------- Forwarded message ---------- From: Nicola Squartini
Date: Tue, Apr 14, 2015 at 9:20 AM Subject: Re: [arch-haskell] gitit status update and why are deps needed for binaries To: Bastien Traverse Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. Right now haskell-pandoc and haskell-gitit are packaged with binaries and modules inside, and the module part depend on all the other packages, including GHC. If we split each of them into two, say pandoc (binaries) and haskell-pandoc (modules), then you could just install the binaries without having to depend on GHC.
The reasons we don't do that sort of splitting are two: 1. The tool that helps with packaging `clbrepo` doesn't support it, and most importantly 2. splitting into -bin and -dev packages, like in Debian/Ubuntu/..., isn't the norm in Arch /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

I am for some kind of splitting. See below. On 14/04/15 06:42, Magnus Therning wrote:
On 14 April 2015 at 02:21, Nicola Squartini
wrote: Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. [..]
The reasons we don't do that sort of splitting are two:
1. The tool that helps with packaging `clbrepo` doesn't support it, and most importantly
Maybe it is a call for looking into such support. Haskell packages are huge! There has to be a significant population of users which want a lean & mean machine. It takes about 900 MiB of installation size for haskell-conduit. Addressing this will help with adoption.
2. splitting into -bin and -dev packages, like in Debian/Ubuntu/..., isn't the norm in Arch
Maybe not -bin and -dev then, but something has to give. I didn't see anything against it in the packaging standards. We can liaise with the core team to find out the Arch-y way for this issue. -- SP

On 14 April 2015 at 13:51, SP
On 14/04/15 06:42, Magnus Therning wrote:
On 14 April 2015 at 02:21, Nicola Squartini
wrote: Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. [..]
The reasons we don't do that sort of splitting are two:
1. The tool that helps with packaging `clbrepo` doesn't support it, and most importantly
Maybe it is a call for looking into such support. Haskell packages are huge! There has to be a significant population of users which want a lean & mean machine. It takes about 900 MiB of installation size for haskell-conduit. Addressing this will help with adoption.
Patches are always welcome. :)
2. splitting into -bin and -dev packages, like in Debian/Ubuntu/..., isn't the norm in Arch
Maybe not -bin and -dev then, but something has to give. I didn't see anything against it in the packaging standards. We can liaise with the core team to find out the Arch-y way for this issue.
We don't really have to include the core team at all. ArchHaskell isn't an official part of Arch so we can do what we want. However, following the path of least surprise would suggest we don't stray too far away from the Arch way. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

Le 14/04/2015 02:20, Nicola Squartini a écrit :
Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. Right now haskell-pandoc and haskell-gitit are packaged with binaries and modules inside, and the module part depend on all the other packages, including GHC. If we split each of them into two, say pandoc (binaries) and haskell-pandoc (modules), then you could just install the binaries without having to depend on GHC.
Thanks, that's the information I was looking for. I hadn't understood that pandoc and gitit could be split in different parts from their regular installation instructions. How does one do so? Le 14/04/2015 14:10, Magnus Therning a écrit :
The reasons we don't do that sort of splitting are two: [...] 2. splitting into -bin and -dev packages, like in Debian/Ubuntu/..., isn't the norm in Arch
Maybe not -bin and -dev then, but something has to give. I didn't see anything against it in the packaging standards. We can liaise with the core team to find out the Arch-y way for this issue.
We don't really have to include the core team at all. ArchHaskell isn't an official part of Arch so we can do what we want. However, following the path of least surprise would suggest we don't stray too far away from the Arch way.
I thought I had read a post from one Arch dev stating splitting in a Debian-like fashion wasn't supported, but I cannot find it right now. The closest I got is the "comparison with Debian" section of the wiki [1] which says:
Arch generally packages software libraries together with their header files, whereas in Debian header files have to be downloaded separately.
It also links to a forum thread exposing the very question of -dev packages in Arch [2], where we can read:
Arch does not split packages the way some other distributions do. Anything with '-dev' at the end will most often be found in the base package.
1. The tool that helps with packaging `clbrepo` doesn't support it, and most importantly
Maybe it is a call for looking into such support. Haskell packages are huge! There has to be a significant population of users which want a lean & mean machine. It takes about 900 MiB of installation size for haskell-conduit. Addressing this will help with adoption.
Patches are always welcome. :)
As much as I'd like to get the very few Haskell packages I need without installing the whole platform, I appreciate Magnus' and Nicola's efforts to maintain the two repos and understand this isn't a priority. I can't provide patches because I'm not a programmer but I'll support such an initiative with what else is needed. [1] https://wiki.archlinux.org/index.php/Arch_compared_to_other_distributions#De... [2] https://bbs.archlinux.org/viewtopic.php?id=179481

On 14 April 2015 at 15:19, Bastien Traverse
Le 14/04/2015 02:20, Nicola Squartini a écrit :
Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. Right now haskell-pandoc and haskell-gitit are packaged with binaries and modules inside, and the module part depend on all the other packages, including GHC. If we split each of them into two, say pandoc (binaries) and haskell-pandoc (modules), then you could just install the binaries without having to depend on GHC.
Thanks, that's the information I was looking for. I hadn't understood that pandoc and gitit could be split in different parts from their regular installation instructions. How does one do so?
IIRC, the regular installation doesn't support it (i.e. via Cabal). What we need is specific recipes to package tools and lib parts into separate packages, i.e. a split package. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

On Tue, Apr 14, 2015 at 10:19 PM, Bastien Traverse
Le 14/04/2015 02:20, Nicola Squartini a écrit :
Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. Right now haskell-pandoc and haskell-gitit are packaged with binaries and modules inside, and the module part depend on all the other packages, including GHC. If we split each of them into two, say pandoc (binaries) and haskell-pandoc (modules), then you could just install the binaries without having to depend on GHC.
Thanks, that's the information I was looking for. I hadn't understood that pandoc and gitit could be split in different parts from their regular installation instructions. How does one do so?
You can pull out /usr/bin/gitit and the other files in /usr/share from the package, and upload only those to your server (once the gitit package is back).

On 14/04/15 13:10, Magnus Therning wrote:
Patches are always welcome. :)
Naturally. If I find time I'll give it a shot.
We don't really have to include the core team at all. ArchHaskell isn't an official part of Arch so we can do what we want. However, following the path of least surprise would suggest we don't stray too far away from the Arch way.
I'm aware and it would be ok to take whatever route suits ArchHaskell best. But given this is targeted at Arch and I presume everyone here wants Haskell to flourish, it might pay dividends in the future to have a design which is in harmony with Arch. Since the design to-date of Arch might have not had to deal with large packages, they may have not catered for this problem and thus no "path of least surprise" for us to follow. It may be worthwhile to have their take on this, a solution they agree is in line with Arch. This way we will also be following a path of least surprise for us, in case they decide in the future to work scenarios like ours differently. Who knows.. in a future where everything is running on Haskell, ArchHaskell might need to be merged in their "extra" repos! -- SP

On Tue, Apr 14, 2015 at 03:41:26PM +0100, SP wrote:
On 14/04/15 13:10, Magnus Therning wrote:
Patches are always welcome. :)
Naturally. If I find time I'll give it a shot.
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus If you can explain how you do something, then you're very very bad at it. -- John Hopfield

On 14/04/15 22:29, Magnus Therning wrote:
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like.
Btw, is there a repo for the PKGBUILD files? -- SP

On 15 April 2015 at 12:28, SP
On 14/04/15 22:29, Magnus Therning wrote:
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like.
Btw, is there a repo for the PKGBUILD files?
No, they are generated, and never checked in. The `cblrepo` database is checked in at https://github.com/archhaskell/habs/ /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

Do the current packages include the dynamically linkable libs? If the apps
are going to require the installation of the Haskell libs they depend on
anyway, maybe having them dynamically link to the Haskell libs, rather than
statically link them in, would save a good chunk of disk space, not to
mention linking time.
On Tue, Apr 14, 2015 at 4:51 AM, SP
I am for some kind of splitting. See below.
On 14/04/15 06:42, Magnus Therning wrote:
On 14 April 2015 at 02:21, Nicola Squartini
wrote: Solving the problem of pandoc and gitit binaries taking so much space, would require splitting the packages in two. [..]
The reasons we don't do that sort of splitting are two:
1. The tool that helps with packaging `clbrepo` doesn't support it, and most importantly
Maybe it is a call for looking into such support. Haskell packages are huge! There has to be a significant population of users which want a lean & mean machine. It takes about 900 MiB of installation size for haskell-conduit. Addressing this will help with adoption.
2. splitting into -bin and -dev packages, like in Debian/Ubuntu/..., isn't the norm in Arch
Maybe not -bin and -dev then, but something has to give. I didn't see anything against it in the packaging standards. We can liaise with the core team to find out the Arch-y way for this issue.
-- SP _______________________________________________ arch-haskell mailing list arch-haskell@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/arch-haskell

On 15 April 2015 at 00:56, Leif Warner
Do the current packages include the dynamically linkable libs? If the apps are going to require the installation of the Haskell libs they depend on anyway, maybe having them dynamically link to the Haskell libs, rather than statically link them in, would save a good chunk of disk space, not to mention linking time.
They do include the shared libs, but the executables don't use them (to my knowledge at least, this of course *has* to be verified first). The last time I looked at using the shared libs there were issues with rpath, basically ghc put in a search path reflecting the build dir and not the install-libdir. This was probably in the 7.6 times though. One would hope this has improved since. The Nix guys did have a solution to this, and IIRC it involved rewriting the rpath after linking. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

Le 14/04/2015 16:12, Magnus Therning a écrit :
IIRC, the regular installation doesn't support it (i.e. via Cabal). What we need is specific recipes to package tools and lib parts into separate packages, i.e. a split package.
Le 14/04/2015 23:29, Magnus Therning a écrit :
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like.
Thanks for having a try at this, it looks good! Have you tested it or you wish for somebody to try it out? Le 15/04/2015 02:15, Nicola Squartini a écrit :
You can pull out /usr/bin/gitit and the other files in /usr/share from the package, and upload only those to your server (once the gitit package is back).
On the other hand that sounds like and excellent idea and very simple to implement (so KISS-compliant) if the modified PKGBUILD-way proves too problematic. I'm thinking of a post-build task packaging the binary along with man pages etc.

On 15 April 2015 at 08:40, Bastien Traverse
Le 14/04/2015 16:12, Magnus Therning a écrit :
IIRC, the regular installation doesn't support it (i.e. via Cabal). What we need is specific recipes to package tools and lib parts into separate packages, i.e. a split package.
Le 14/04/2015 23:29, Magnus Therning a écrit :
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like.
Thanks for having a try at this, it looks good! Have you tested it or you wish for somebody to try it out?
I have not tested it, and there might be quite a bit of testing necessary I'm afraid. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

I think it's unnecessary (and even annoying for many users) to have all
packages that contains binaries split. Beside it requires modifying cblrepo.
Unless the demand increases, I would suggest to simply manually split (via
*.pkgbuild patches) only explicitly requested packages, like pandoc and
gitit.
On Wed, Apr 15, 2015 at 7:15 PM, Magnus Therning
On 15 April 2015 at 08:40, Bastien Traverse
wrote: Le 14/04/2015 16:12, Magnus Therning a écrit :
IIRC, the regular installation doesn't support it (i.e. via Cabal). What we need is specific recipes to package tools and lib parts into separate packages, i.e. a split package.
Le 14/04/2015 23:29, Magnus Therning a écrit :
Just playing around a little with one of the packages currently in the repo comprising both a lib and a binary resulted in the attached PKGBUILD (shake). It might be close to what a solution could look like.
Thanks for having a try at this, it looks good! Have you tested it or you wish for somebody to try it out?
I have not tested it, and there might be quite a bit of testing necessary I'm afraid.
/M
-- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus _______________________________________________ arch-haskell mailing list arch-haskell@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/arch-haskell

On 15 April 2015 at 12:28, Nicola Squartini
I think it's unnecessary (and even annoying for many users) to have all packages that contains binaries split. Beside it requires modifying cblrepo. Unless the demand increases, I would suggest to simply manually split (via *.pkgbuild patches) only explicitly requested packages, like pandoc and gitit.
I tend to agree, but this has come up a few times now, so if someone's interested in putting in the time for a robust patch then I'll be happy to test it. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

Still on the topic of reducing installation dependencies: are self-contained binaries in the style of pandoc [1] doable for gitit and other Haskell packages as well? This could be an interesting way to address the issue.
It is possible to compile pandoc such that the data files pandoc uses are embedded in the binary. (The executables in the binary are built this way.) The resulting binary can be run from any directory and is completely self-contained.
cabal update cabal install hsb2hs cabal install --flags="embed_data_files" pandoc pandoc-citeproc
[1] http://pandoc.org/installing.html#creating-a-relocatable-binary

No gitit doesn't have that flag.
Even in the case of pandoc, it will not make your installation smaller,
just embed the support files (/usr/share I guess) inside the
/usr/bin/pandoc binary.
On Wed, Apr 15, 2015 at 6:49 PM, Bastien Traverse
Still on the topic of reducing installation dependencies: are self-contained binaries in the style of pandoc [1] doable for gitit and other Haskell packages as well? This could be an interesting way to address the issue.
It is possible to compile pandoc such that the data files pandoc uses are embedded in the binary. (The executables in the binary are built this way.) The resulting binary can be run from any directory and is completely self-contained.
cabal update cabal install hsb2hs cabal install --flags="embed_data_files" pandoc pandoc-citeproc
[1] http://pandoc.org/installing.html#creating-a-relocatable-binary _______________________________________________ arch-haskell mailing list arch-haskell@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/arch-haskell

The last time I looked at using the shared libs there were issues with rpath, basically ghc put in a search path reflecting the build dir and not the install-libdir. This was probably in the 7.6 times though. One would hope this has improved since. The Nix guys did have a solution to this, and IIRC it involved rewriting the rpath after linking.
A more or less elegant solution to the rpath problem was found in a previous thread [1]: Using GHC `-dynload=deploy' flag [3] and a `/etc/ld.so.conf.d/haskell.conf' file which specify the location of the shared libraries, e.g. `/usr/lib/ghc-7.6.3/sharedg'. Details in [2]. This solution was working well for most of the package, I remember being able to build pandoc with shared libraries, but sadly not for gtk, which has a kind of bootstrapping method that was not working with `-dynload=deploy' flag. It was with GHC 7.6, but as show in [3] the option is still present in latest GHC. Maybe it is worth another try, gtk could have changed! Not sure to have the time, but if yes I'll try it again. ++ Fabien [1] http://comments.gmane.org/gmane.comp.lang.haskell.arch-linux/1792 [2] http://permalink.gmane.org/gmane.comp.lang.haskell.arch-linux/1795 [3] https://downloads.haskell.org/~ghc/latest/docs/html/users_guide/using-shared...

On 15 April 2015 at 12:57, Fabien Dubosson
The last time I looked at using the shared libs there were issues with rpath, basically ghc put in a search path reflecting the build dir and not the install-libdir. This was probably in the 7.6 times though. One would hope this has improved since. The Nix guys did have a solution to this, and IIRC it involved rewriting the rpath after linking.
A more or less elegant solution to the rpath problem was found in a previous thread [1]: Using GHC `-dynload=deploy' flag [3] and a `/etc/ld.so.conf.d/haskell.conf' file which specify the location of the shared libraries, e.g. `/usr/lib/ghc-7.6.3/sharedg'. Details in [2].
This solution was working well for most of the package, I remember being able to build pandoc with shared libraries, but sadly not for gtk, which has a kind of bootstrapping method that was not working with `-dynload=deploy' flag.
It was with GHC 7.6, but as show in [3] the option is still present in latest GHC. Maybe it is worth another try, gtk could have changed! Not sure to have the time, but if yes I'll try it again.
Ah, yes, that's right, it worked even better than I remembered... and gtk was the culprit that threw a spanner in the works. It might be worth a try now, no matter where the issue of package splitting goes. /M -- Magnus Therning OpenPGP: 0xAB4DFBA4 email: magnus@therning.org jabber: magnus@therning.org twitter: magthe http://therning.org/magnus

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 On 15/04/15 11:57, Fabien Dubosson wrote:
A more or less elegant solution to the rpath problem was found in a previous thread [1]: Using GHC `-dynload=deploy' flag [3] and a `/etc/ld.so.conf.d/haskell.conf' file which specify the location of the shared libraries, e.g. `/usr/lib/ghc-7.6.3/sharedg'. Details in [2].
Thanks for all this background info Fabien, it helps get this going.
This solution was working well for most of the package, I remember being able to build pandoc with shared libraries, but sadly not for gtk, which has a kind of bootstrapping method that was not working with `-dynload=deploy' flag.
One package shouldn't warp the whole package system. One should push suggestions upstream and they should be accepted if there is technical merit. In the end there is always the patch approach for rogues. - -- SP

Thanks for all this background info Fabien, it helps get this going.
My pleasure!
One package shouldn't warp the whole package system. One should push suggestions upstream and they should be accepted if there is technical merit. In the end there is always the patch approach for rogues.
I agree. As far as I remember Magnus an me were both out of time to look further into this. Maybe things have changed since. I started rebasing my past work on the last version of habs, I'll write another mail soon for a status. ++ Fab

On 15/04/15 21:44, Fabien Dubosson wrote:
I agree. As far as I remember Magnus an me were both out of time to look further into this. Maybe things have changed since. I started rebasing my past work on the last version of habs, I'll write another mail soon for a status.
Ok. I was looking for branches relating to this feature. I noticed you have one, but need to find one which has the changes proposed for cblrepo. Either way I will try to put something together, but chances are my snail pace will get beaten! Keep me posted either directly or maybe new thread? Also, been looking to see what makes things so large in the filebase. Just checking in ghc's dir, it's static libraries: ``` 185480024 /usr/lib/ghc-7.10.1/ghc_EMlWrQ42XY0BNVbSrKixqY/libHSghc-7.10.1-EMlWrQ42XY0BNVbSrKixqY_p.a 109044692 /usr/lib/ghc-7.10.1/ghc_EMlWrQ42XY0BNVbSrKixqY/libHSghc-7.10.1-EMlWrQ42XY0BNVbSrKixqY.a 76260686 /usr/lib/ghc-7.10.1/Cabal_HWT8QvVfJLn2ubvobpycJY/libHSCabal-1.22.2.0-HWT8QvVfJLn2ubvobpycJY_p.a 67111580 /usr/lib/ghc-7.10.1/site-local/highlighting-kate-0.5.14/libHShighlighting-kate-0.5.14-5S9jZ1dVOgcGI7XstIXNsq_p.a 63048936 /usr/lib/ghc-7.10.1/ghc_EMlWrQ42XY0BNVbSrKixqY/libHSghc-7.10.1-EMlWrQ42XY0BNVbSrKixqY-ghc7.10.1.so 57385398 /usr/lib/ghc-7.10.1/Cabal_HWT8QvVfJLn2ubvobpycJY/libHSCabal-1.22.2.0-HWT8QvVfJLn2ubvobpycJY.a 46972990 /usr/lib/ghc-7.10.1/base_I5BErHzyOm07EBNpKBEeUv/libHSbase-4.8.0.0-I5BErHzyOm07EBNpKBEeUv_p.a 35159482 /usr/lib/ghc-7.10.1/base_I5BErHzyOm07EBNpKBEeUv/libHSbase-4.8.0.0-I5BErHzyOm07EBNpKBEeUv.a 27632384 /usr/lib/ghc-7.10.1/site-local/texmath-0.8.0.2/libHStexmath-0.8.0.2-1UuTXL14IBUBNFTO6p9bK5_p.a 25808360 /usr/lib/ghc-7.10.1/site-local/highlighting-kate-0.5.14/libHShighlighting-kate-0.5.14-5S9jZ1dVOgcGI7XstIXNsq.a ``` -- SP
participants (6)
-
Bastien Traverse
-
Fabien Dubosson
-
Leif Warner
-
Magnus Therning
-
Nicola Squartini
-
SP