Re: [boost] [build] bootstrap.sh is still broken

Subject: Re: [boost] [build] bootstrap.sh is still broken
From: Stephan T. Lavavej (stl_at_[hidden])
Date: 2013-10-27 16:50:02

[Eric Niebler] 
> Committed to trunk, https://svn.boost.org/trac/boost/changeset/86460.
> Stephan, can you have a look at the changeset and let me know if
> everything looks ok?

Yay, thank you! The diff looks good to me.

> I don't have MinGW installed. I'd love it if someone tried:
> > $ ./bootstrap.sh --with-toolset=mingw
> ...before I merge this to release.

I believe this will still explode due to the first problem I identified -
different parts of Boost.Build can't agree on whether MinGW should be called
gcc or mingw. Last I checked, no permutation of command-line arguments to
bootstrap.sh and b2.exe could avoid this.

[niXman] 
> What I am doing wrongly?

One of the problems appears to be that you're attempting to build
Boost.Context with only MinGW installed (no VC). This is broken due to
https://svn.boost.org/trac/boost/ticket/7262 . Context's maintainer has
cited a MinGW linker bug and has refused to work around it. (I don't know
whether this bug has been reported to binutils.) Boost.Build does not
auto-detect this situation and disable Context, so you must pass
--without-context --without-coroutine in order to avoid this (Coroutine
depends on Context).

Additionally, if you are building with mingw-w64, you will want to pass
pch=off in order to avoid having Boost.Math trigger a mingw-w64 PCH bug
(unless it has been fixed in the last month or so). See
http://lists.boost.org/Archives/boost/2013/07/205244.php for that thread.

I don't know what's happening with the static assertion you're getting in
Boost.Math. That might have appeared since 1.54.0, in which case I wouldn't
have seen it yet.

STL