New Rtools is GCC 4.9.2!

[Daniel Lee]

See forwarded message from Avraham!

---------- Forwarded message ----------
From: Avraham Adler <avraha...@gmail.com>
Date: Wed, Mar 11, 2015 at 5:26 PM
Subject: New Rtools is GCC 4.9.2!
To: Daniel Lee <bea...@alum.mit.edu>, Bob Carpenter <ca...@alias-i.com>

This thread on r-devel is a bit long <https://stat.ethz.ch/pipermail/r-devel/2015-March/070785.html> but the latest Rtools is based on GCC 4.9.2, and R builds properly, so when 3.2 is released in April, you can probably start taking advantage of cool C++11 stuffs.

I doubt that the default binary of R will be built with std=c++11 or gnu++11, but most people running stan should be able to compile R as they need to for stand anyway, no?

Avi

[Ben Goodrich]

That is good news. Allen, I seem to remember the last time you wrote about C++11 on Windows that whatever was the default Anaconda version on the 3.x Python series was built with VS2013 or something unproblematic. Is that still the case? If we're good there, then I think we should switch over to the C++11 standard after R 3.2.0 is released. We will have to drop support for the old llvm-g++-4.2 compiler on Macs but I think the gain would be worth it.

Ben

[Bob Carpenter]

Being able to use C++11 would be great. I started a wiki
page a while back to track C++11 issues:

https://github.com/stan-dev/stan/wiki/Cpp11-Upgrade

Allen said that for Windows/Python, we need to restrict
ourselves to the C++ features in Visual Studio 2010.

It would be soooo nice if we could just abandon Windows.
It's been a huge thorn in our side from the get-go.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Allen B. Riddell]

Hi Ben,

Yes, that's right. The only real concern is that we'll likely lose
Python 2.7 users. There are, alas, many Python 2.7 users. But they can
just use the pre-C+11 version of Stan or switch to Python 3.

For their sake, I think it would be good to announce publicly what
version will turn on C++11 features and maybe go point release cycle
(2.7?)---or at least a few months---with a visible deprecation warning
before turning C+11 on.

Best,

Allen

[Bob Carpenter]

Whatever we do, I do not want to have to maintain
multiple versions of Stan. Our releases and mailing list
issues are complicated enough already.

I'm not sure what you mean by "point release cycle" or
what we should be announcing as deprecated.

- Bob

[Allen B. Riddell]

Any sort of public time frame would be fine. e.g., switch in 3 months, 6
months, or Stan 2.8 etc..

I'll be adding a deprecation warning that only shows up for Python 2.7
users of PyStan.

Best,

-a

[Avraham Adler]

Unfortunately, there are some serious hiccups with the new chain and Rcpp and some other packages. For example, see this thread <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/2015-March/008489.html> and this thread <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/2015-March/008495.html>. As Stan uses Rcpp, this may be an issue, and regardless, Dr. Duncan Murdoch (keeper of Rtools) may revert back to Rtools 3.2 if this cannot be figured out quickly.

Avi

[Avraham Adler]

FYI for those not subscribed to Rcpp-devel: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/2015-March/008500.html>. Key statement by Dr. Murdoch:


*****Begin Quote*****
I think there are 2 likely choices, with a third possible. We could use the Rtools33 toolchain that is currently on CRAN. This appears likely to require changes by C++ users. Alternatively, we could revert to the previous toolchain, based on gcc 4.6.3. If we do revert, we will be able to change for R 3.2.1, which is unscheduled, but likely to appear in June, judging by past history.

The 3rd possibility is that someone other than me makes small changes to
the toolchain (the scripts are on CRAN in bin/windows/Rtools/scripts),
and enough people test the resulting chain so we have general agreement
that the modification works. It needs to be someone other than me,
because I need to try it out in my own tests, and that means I need to
have it in hand by Wednesday, if I'm going to decide by Friday. I will
not build this into Rtools before next Friday, so you'll need to
distribute this by some other means.
*****End Quote*****

Avi

[Allen B. Riddell]

As I dig into this I'm realizing that it might only be Python 2.7
Windows users who will be out of luck with C++11. My Python 2.7 on
Ubuntu/Debian Linux is compiled with gcc 4.8.2 -- so I suspect it will
continue to work.

Best,

-a

On 03/12, Bob Carpenter wrote: