So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
http://hg.et.redhat.com/misc/fedora-mingw--devel/ (click 'manifest' then 'ocaml' or the 'ocaml-*' subdirectories) mingw32-ocaml - the cross-compiler (native Fedora prog) mingw32-ocaml-calendar - cross-compiled OCaml Calendar library mingw32-ocaml-csv - cross-compiled OCaml CSV library mingw32-ocaml-curses - cross-compiled OCaml curses bindings mingw32-ocaml-extlib - cross-compiled OCaml extlib (library) mingw32-ocaml-findlib - (this just has a single config file) mingw32-ocaml-lablgl - cross-compiled OCaml OpenGL bindings mingw32-ocaml-lablgtk - cross-compiled OCaml Gtk bindings mingw32-ocaml-libvirt - cross-compiled OCaml libvirt bindings mingw32-ocaml-xml-light - cross-compiled OCaml XML library
When we were doing the original packaging, we explored the option of building the MinGW packages as subpackages of the native Fedora packages, eg: the gnutls specfile would contain the '-n mingw32-gnutls' subpackage. This would have made it simpler to keep the packages in synch with native Fedora, but we felt that native packagers wouldn't be so happy about this. They wouldn't be expert in MinGW packaging, and would drop the MinGW subpackage at the first sign of trouble.
However, since I'm in the unique position of maintaining most of the corresponding ocaml-* (native) and mingw32-ocaml-* (cross-compiled) packages, I would like to ask whether I can create the packages above and others in future as subpackages of the already approved ocaml-* native packages.
I'm not looking for any exception to the relevant packaging guidelines, which I would try to follow (excepting any mistakes).
If you want to see what some RPMs look like, see: http://www.annexia.org/tmp/mingw/fedora-10/x86_64/RPMS/
Thoughts? I set follow-ups to the fedora-packaging list.
Rich.
Richard W.M. Jones wrote:
On Fri, Dec 05, 2008 at 11:12:25PM +0000, Richard W.M. Jones wrote:
So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
Everyone's happy with me to go ahead with this?
I'm not but I've been involved with other arguments so I haven't been able to think about whether this is just an unjustifiable gut feeling or if there's an actual reason this should be avoided.
On Mon, 2008-12-08 at 17:53 +0000, Richard W.M. Jones wrote:
On Fri, Dec 05, 2008 at 11:12:25PM +0000, Richard W.M. Jones wrote:
So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
Everyone's happy with me to go ahead with this?
I am not, but I am sure if I understand you correctly.
Is your plan to build mingw packages as part of building Fedora packages within the Fedora build system? If yes, I am opposed to doing this.
Is your plan to add support to your ocaml package to enable you to reuse your ocaml Fedora rpm.specs to build mingw package outside of Fedora? I'd consider this to be "your private pleasures" and to be "outside of Fedora's business".
Ralf
On Tue, 2008-12-09 at 07:45 +0100, Ralf Corsepius wrote:
On Mon, 2008-12-08 at 17:53 +0000, Richard W.M. Jones wrote:
On Fri, Dec 05, 2008 at 11:12:25PM +0000, Richard W.M. Jones wrote:
So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
Everyone's happy with me to go ahead with this?
I am not, but I am sure if I understand you correctly.
.... not sure ..., sorry
Is your plan to build mingw packages as part of building Fedora packages within the Fedora build system? If yes, I am opposed to doing this.
Is your plan to add support to your ocaml package to enable you to reuse your ocaml Fedora rpm.specs to build mingw package outside of Fedora? I'd consider this to be "your private pleasures" and to be "outside of Fedora's business".
On Tue, Dec 09, 2008 at 07:45:45AM +0100, Ralf Corsepius wrote:
On Mon, 2008-12-08 at 17:53 +0000, Richard W.M. Jones wrote:
On Fri, Dec 05, 2008 at 11:12:25PM +0000, Richard W.M. Jones wrote:
So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
Everyone's happy with me to go ahead with this?
I am not, but I am sure if I understand you correctly.
Is your plan to build mingw packages as part of building Fedora packages within the Fedora build system? If yes, I am opposed to doing this.
I'm a bit confused by this. You are aware that we are planning to build lots of mingw packages in Fedora (using the build system) to enable cross-compilation? This has been discussed to death already, and approved, so it shouldn't be news to anyone.
http://fedoraproject.org/wiki/MinGW http://fedoraproject.org/wiki/Features/Windows_cross_compiler http://fedoraproject.org/wiki/Packaging/MinGW https://bugzilla.redhat.com/buglist.cgi?version=&component=&target_m...
Is your plan to add support to your ocaml package to enable you to reuse your ocaml Fedora rpm.specs to build mingw package outside of Fedora? I'd consider this to be "your private pleasures" and to be "outside of Fedora's business".
No, I would like to build the cross-compiler from the ocaml.spec, instead of doing a separate review request for a new mingw32-ocaml package. (Plus the libraries too as mentioned in the original email).
Rich.
On Tue, 2008-12-09 at 10:52 +0000, Richard W.M. Jones wrote:
On Tue, Dec 09, 2008 at 07:45:45AM +0100, Ralf Corsepius wrote:
On Mon, 2008-12-08 at 17:53 +0000, Richard W.M. Jones wrote:
On Fri, Dec 05, 2008 at 11:12:25PM +0000, Richard W.M. Jones wrote:
So as I mentioned before we have an OCaml Windows cross-compiler as part of the Fedora MinGW project:
Everyone's happy with me to go ahead with this?
I am not, but I am sure if I understand you correctly.
Is your plan to build mingw packages as part of building Fedora packages within the Fedora build system? If yes, I am opposed to doing this.
I'm a bit confused by this. You are aware that we are planning to build lots of mingw packages in Fedora (using the build system) to enable cross-compilation?
No, but if so, I have to reiterate to be vehemently disagreeing with this and will never agree with it.
Shipping cross-toolchains as part of Fedora to enable users to build MinGW binaries is OK with me, but abusing Fedora to support Windows is not OK with me.
Ralf
"RWMJ" == Richard W M Jones rjones@redhat.com writes:
RWMJ> Everyone's happy with me to go ahead with this?
You have to torture reality pretty badly to assume that silence somehow implies happiness. Maybe people are just annoyed to see this come up again.
So here's one from someone who was never all that comfortable with all of this Windows support in the first place:
During the discussions you were somewhat adamant about the limited number of Windows library packages that your proposal would entail. You quoted numbers and relatively small size requirements as evidence that this was no big deal. Now you're talking about taking what is something of a niche package category (ocaml packages) and adding a second level of niche-itude to them (windows cross-compilation environment for ocaml packages) and I'm wondering if the overhead of these packages was included in your initial figures and whether you actually think that anyone other than you will actually use them.
I mean, sure, if you're a packager and you can get someone to review your packages, you can basically turn Fedora into your own personal distro, with the specialized packages that you want already in there. I don't think that's a bad thing. Even better if other people happen to benefit from those packages. At some point, however, someone needs to actually think about how the cost of this compares to the benefits.
I do have a couple of other hands in the fray, though:
As a package reviewer, I think you've already dropped a metric ass-ton of packages on the review queue and I shudder to think that you would consider actually adding more without spending at least a solid month helping us review packages. Avoiding having to review another pile of whatever-for-windows packages would be great.
As a Packaging Committee member, I would want you to at least add sufficient comments to these specfiles to discourage anyone who might want to package an ocaml module from using them as examples unless they somehow want to maintain them for Windows as well. After taht long review process we have what I think are a good set of ocaml package guidelines with nice templates, and now you're proposing to take the bulk of those packages away from that.
- J<
On Tue, Dec 09, 2008 at 09:41:41AM -0600, Jason L Tibbitts III wrote:
"RWMJ" == Richard W M Jones rjones@redhat.com writes:
RWMJ> Everyone's happy with me to go ahead with this?
You have to torture reality pretty badly to assume that silence somehow implies happiness. Maybe people are just annoyed to see this come up again.
I was a bit surprised there was no response from the first email, and the second email was an opportunity to bring this to peoples' attention again. I haven't made any changes yet to the packages, precisely because I wanted to hear what people had to say and ensure there was general agreement first.
So here's one from someone who was never all that comfortable with all of this Windows support in the first place:
During the discussions you were somewhat adamant about the limited number of Windows library packages that your proposal would entail. You quoted numbers and relatively small size requirements as evidence that this was no big deal.
This is still the case. Even with all the packages we've done, plus the OCaml subpackages, the whole of mingw is under a gigabyte (src + noarch RPMs). At the time we estimated 800 MB, so we have gone slightly over our estimate, but at the same time we've expanded the scope to include C++ libraries like gtkmm. Still, I wouldn't really say that 1 GB is excessive.
We have found a non-virt co-maintainer (Levente Farkas) for all the base packages too.
Now you're talking about taking what is something of a niche package category (ocaml packages) and adding a second level of niche-itude to them (windows cross-compilation environment for ocaml packages) and I'm wondering if the overhead of these packages was included in your initial figures and whether you actually think that anyone other than you will actually use them.
Yes, people in the OCaml community are excited by this. That may not be a community that is very visible to Fedora packagers I admit.
I mean, sure, if you're a packager and you can get someone to review your packages, you can basically turn Fedora into your own personal distro, with the specialized packages that you want already in there. I don't think that's a bad thing. Even better if other people happen to benefit from those packages. At some point, however, someone needs to actually think about how the cost of this compares to the benefits.
I do have a couple of other hands in the fray, though:
As a package reviewer, I think you've already dropped a metric ass-ton of packages on the review queue and I shudder to think that you would consider actually adding more without spending at least a solid month helping us review packages. Avoiding having to review another pile of whatever-for-windows packages would be great.
It's true that I have been negligent in doing very little review work. So I will try to change that.
As a Packaging Committee member, I would want you to at least add sufficient comments to these specfiles to discourage anyone who might want to package an ocaml module from using them as examples unless they somehow want to maintain them for Windows as well. After taht long review process we have what I think are a good set of ocaml package guidelines with nice templates, and now you're proposing to take the bulk of those packages away from that.
OK, this is fair too. I usually encourage OCaml packagers to start out with:
http://fedoraproject.org/wiki/Image:Packaging_OCaml_ocaml-foolib.spec
linked from:
http://fedoraproject.org/wiki/Packaging/OCaml
That doesn't currently include a mingw subpackage, and nor should it.
Note that there are currently 70 OCaml packages in Fedora, and only a handful of those (10) were proposed to be cross-compiled, 2 of those being the cross-compiler itself.
Rich.
Richard W.M. Jones wrote:
OK, this is fair too. I usually encourage OCaml packagers to start out with:
http://fedoraproject.org/wiki/Image:Packaging_OCaml_ocaml-foolib.spec
linked from:
http://fedoraproject.org/wiki/Packaging/OCaml
That doesn't currently include a mingw subpackage, and nor should it.
Note that there are currently 70 OCaml packages in Fedora, and only a handful of those (10) were proposed to be cross-compiled, 2 of those being the cross-compiler itself.
So.. I'd like to stick with the current Guidelines that require separate reviews and packages for cross-compiled packages. Here's some of the logic:
1) Just because you are the maintainer of the package now doesn't mean that you'll be the maintainer in Fedora 27.
2) Rather than special-case OCaml, (or really... the OCaml subset maintained by you) it would be better to write a general rule. However, that general rule leads to a bit of confusion -- subpackages may be used when the maintainer is amenable to it. Separate packages when the maintainer is not. And when maintainership switches, go with whatever was done before? Or merge and drop as the current maintainer wishes? This is all very ugly.
3) Since we currently review packages on import time (instead of having an ongoing re-review process) and cross-compilation is a very large spec to add to the build, it's better to make sure cross-compilation is reviewed via separate packages than to have it added on to random packages willy-nilly.
4) Having cross-compiled sub-packages prevents us from separating out the cross-compiled packages should that become desirable in the future (I sit on Infrastructure and FPC and watch FESCo so I know how difficult this is from the infrastructure side and yet how desired it was from the other.) Once again, this is keeping options open for Fedora 27.
-Toshio
On Tue, Dec 09, 2008 at 10:46:08AM -0800, Toshio Kuratomi wrote:
Richard W.M. Jones wrote:
OK, this is fair too. I usually encourage OCaml packagers to start out with:
http://fedoraproject.org/wiki/Image:Packaging_OCaml_ocaml-foolib.spec
linked from:
http://fedoraproject.org/wiki/Packaging/OCaml
That doesn't currently include a mingw subpackage, and nor should it.
Note that there are currently 70 OCaml packages in Fedora, and only a handful of those (10) were proposed to be cross-compiled, 2 of those being the cross-compiler itself.
So.. I'd like to stick with the current Guidelines that require separate reviews and packages for cross-compiled packages. Here's some of the logic:
- Just because you are the maintainer of the package now doesn't mean
that you'll be the maintainer in Fedora 27.
- Rather than special-case OCaml, (or really... the OCaml subset
maintained by you) it would be better to write a general rule. However, that general rule leads to a bit of confusion -- subpackages may be used when the maintainer is amenable to it. Separate packages when the maintainer is not. And when maintainership switches, go with whatever was done before? Or merge and drop as the current maintainer wishes? This is all very ugly.
- Since we currently review packages on import time (instead of having
an ongoing re-review process) and cross-compilation is a very large spec to add to the build, it's better to make sure cross-compilation is reviewed via separate packages than to have it added on to random packages willy-nilly.
- Having cross-compiled sub-packages prevents us from separating out
the cross-compiled packages should that become desirable in the future (I sit on Infrastructure and FPC and watch FESCo so I know how difficult this is from the infrastructure side and yet how desired it was from the other.) Once again, this is keeping options open for Fedora 27.
This seems very reasonable. So I'll introduce these as new packages instead of doing them as subpackages. (And try to do some more reviews ...)
Rich.
packaging@lists.fedoraproject.org
-
Jason L Tibbitts III -
Ralf Corsepius -
Richard W.M. Jones -
Toshio Kuratomi