Re: One-Click Installer: MinGW? or VC2005?

On 7/20/06, Kaspar Schiess <[e-mail]> wrote:
>
> Dear Curt,
>
> I have been a silent reader of your posts about the impe(n)ding decision
> you have to take for either the MS VC++ or the mingw toolchain.
>
> Let's just sum up what has been said/can be said:
>
> - We need to go down either road *completely*. There seems to be no
> middle way. Or we could go down both roads, but would get stuck halfway
> (extension X being available for mswin builds, but not mingw builds,
> extension Y for mingw and not mswin, which to choose?).
>
> - Mingw certainly has less documentation, but puts us in 'control' of
> availability of compilers.
>
> - Microsoft seems to like the thought of you choosing MS VC++ publicly
> and assures you of their support.
>
> - Ara Howard makes a good point by saying that unless a library (ruby
> extension or not) is explicitly constructed to build under MS VC++, it
> will
> require a lot of fiddling to get it built. The MS toolchain is just ()
> very
> different from a unix toolchain () inadequate (choose what you prefer). I
> have myself chosen mingw over MS VC++ because of that.
>
> - The same point can be made pro MS VC++ by saying that often,
> compilation of unix libs on mingw requires fiddling (and very unixish
> fiddling) too. I would wholly agree.
>
> - The vision a lot of people have is to bring us closer to the
> installers-only universe that python has successfully created. Only that
> we
> are divided between the two compilers. We risk loosing momentum on a wrong
> decision.


Thanks for the conscise summary!

I really can't add to that; I think your decision is a hard one to make and
> there isn't a concise list of pros and cons to chose from. I (personally
> and as the maintainer for RMagick windows build) would be ready to invest
> my time in the following setups:
>
> a. A 'pure' mingw build. This would be the best option for someone like
> me who likes using unix tools.
>
> b. A compiler farm setup with either mingw or MSVC++. We should be able
> to deal out logins / distribute virtual disk images. This setup would have
> to be maintained. Everything needs to be compiled there / using that
> virtual machine. Requires close collaboration and some funding.
>
> c. Going down both ways. Requires more manpower and may put us in
> either/or situations further along. RMagick would probably use mingw in
> this setup. Note that I currently know of no one that has succeded an
> RMagick build on a MS VC++ setup; but I am sure that this could be fixed
> given the proper time investment.
>
> I am grateful that you take the time to ask these questions. I hope I have
> advanced the discussion.
>
> best regards,
> Kaspar Schiess


The manpower issue is very real. Its difficult enough to find help with one
compiler -- two would be impossible. MinGW is still looking like the way to
go, although I'm not yet ready to declare that without a little more due
consideration.

There is one major hole in the build processes for the one-click installer
that I want to fix along with this compiler transition -- tests for the
included extensions -- there aren't any. This needs to change. For each
included extension there needs to be at least a few basic automated tests to
ensure that the extension is functioning properly. If anyone wants to
volunteer to help with this, please email me (curt at hibbs dot com) and
I'll hold your name in reserve until I get to that point.

Curt