Re: Mingw 64 bit build

Markus Wanner <[email protected]> writes:

> I understand your point of view and appreciate your efforts. Please
> continue to maintain the msys2 documentation.

Ok, thanks.

> I'm not quite satisfied with the general guide for Windows, though, so
> please allow me to write something. I think that may well fit into
> INSTALL itself.

Ok.

> Do you have any useful hints for what build system to choose for
> Windows? I mean Cygwin vs the others is somewhat obvious - either you
> want that POSIX emulation layer or you don't. 

Right.

> But all of the MinGW versions confuse me a lot. Why did you choose
> MinGW64 over MinGW-w64 (!) [0], for example? 

There are a lot of ways to use MinGW64; it is confusing.

I did not try them all; I only tried downloading from the MinGW64 site,
and the Msys2 site.

I went with Msys2 because it has a good package manager, an almost
complete set of packages for monotone (only Botan needed to be compiled
from source), can run autotools and configure, and a helpful email list.

> What about the different C++ exception and threading models? Which one
> did you (or mingw/msys2) use? 

It was chosen by Msys2 when they packaged MinGW64; I don't know which
they picked. It should be possible to find out from the Msys2 project.
But at this point, I also don't care :).

> What effect do these options possibly have on monotone? 

No idea, except it's pretty minor. We use exceptions for error handling,
not for mainline processing.

> Let alone the issue of cross-compiling 32-bit binaries from a 64-bit
> system & tool chain. 

The tool chain installed in msys2/mingw32 is a 32 bit toolchain.

At the moment, it Just Works, so I'm happy.

> (And vice versa?) And then there is also MSVC...

I've never installed MSVC; I see no reason to. But apparently some
people do: more power to them.

> Of course, we cannot ever test all possible combinations. We don't do
> that on any Unix, either. But rather than listing just one or two very
> specific combinations, we can still state the options, their
> requirements, what's known to work or fail (for example the issue you
> faced with gcc 4.6.2).

I prefer to just list the one known working config; that's hard enough.
No one has ever asked for anything more (until you, now).

> FWIW, I also plan to keep my buildbot on msys 1.0, for now. While I
> don't intend to maintain the INSTALL_windows_mingw.txt document to the
> level of detail you used to do it, I'd still like to keep something in
> there that clarifies that MinGW / Msys 1.0 is known to work. Dropping
> the file, as you suggested, would lose that knowledge.

It didn't work for me, so "known to work" is not true, I think. It would
be best if more than one person was able to follow the install script
and succeed.

By that standard, msys2 isn't "known to work", either :(.

Cygwin also didn't work for me. 

Sometimes it's just that no two Windows boxes are ever the same; it is
frustrating.

> I'll eventually write up something.

Ok.

-- 
-- Stephe