search

alashworth / test-issue-import

0 stars 0 forks source link

Fix Windows integration tests. #115

Open alashworth opened 5 years ago

alashworth commented 5 years ago

Issue by syclik Wednesday Sep 07, 2016 at 14:46 GMT Originally opened as https://github.com/stan-dev/stan/issues/2058

Summary:

Integration tests fail on Windows.

Description:

With the change to the integration tests trying to compile every model, we've now broken our Windows tests builds. It's trying to create executables that are too big for the linker using RTools's g++ 4.6.3.

I don't know if this problem no longer exists with a newer version of RTools.

The fix is to break apart the Stan programs into smaller chunks. There's a limit to the size of a Stan program and its different C++ instantiations under Windows. I don't think any user will realistically hit this limit unless they're generating code.

Reproducible Steps:

Run:

make src/test/test-models/good/function-signatures/distributions/univariate/continuous/exp_mod_normal/exp_mod_normal_ccdf_log_1.hpp-test

There are a few more that will fail. That's the first failure.

Current Output:

...
c:/rtools/gcc-4.6.3/bin/../lib/gcc/i686-w64-mingw32/4.6.3/../../../../i686-w64-mingw32/bin/as.exe: nul: too many sections (35710)
{standard input}: Assembler messages:
{standard input}: Fatal error: can't write nul: File too big

Expected Output:

It should just pass.

Additional Information:

Maybe our tests on Windows should just skip this?

Current Version:

v2.12.0

alashworth commented 5 years ago

Comment by aadler Wednesday Sep 07, 2016 at 16:41 GMT

Perhaps it's time to make rstan depend on R >= 3.3 and only use the 4.9.3 version of Rtools?

alashworth commented 5 years ago

Comment by syclik Thursday Sep 08, 2016 at 04:03 GMT

I'd prefer to have minimal requirements wherever possible. Here, we're just pushing the ability of the 32-bit linker that comes with the compiler, I think. If you look at what it's trying to do, it's not really realistic. There's no need for us to force users to upgrade just to keep up with tests that aren't even run by users.

On Wed, Sep 7, 2016 at 12:41 PM, Avraham Adler notifications@github.com wrote:

Perhaps it's time to make rstan depend on R >= 3.3 and only use the 4.9.3 version of Rtools?

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stan-dev/stan/issues/2058#issuecomment-245342475, or mute the thread https://github.com/notifications/unsubscribe-auth/AAZ_Fyn9K-uNhP0PfyIPWIFdFCB2tYCAks5qnuk6gaJpZM4J3AMe .

alashworth commented 5 years ago

Comment by bob-carpenter Thursday Mar 30, 2017 at 20:34 GMT

@sycklik Is this still a problem? I added Sean as an asignee.

I'm also OK just skipping integration tests on Windows assuming we're doing it elsewhere if Windows can't handle it and this is going to be some hugely time-consuming process to fix. Otherwise, I'm just bumping to 2.15++ now until a decision can be made.

alashworth commented 5 years ago

Comment by syclik Thursday Mar 30, 2017 at 20:51 GMT

Yes, this is still an outstanding issue. But I'm happy to give up on it for now.

On Thu, Mar 30, 2017 at 4:34 PM, Bob Carpenter notifications@github.com wrote:

@sycklik Is this still a problem? I added Sean as an asignee.

I'm also OK just skipping integration tests on Windows assuming we're doing it elsewhere if Windows can't handle it and this is going to be some hugely time-consuming process to fix. Otherwise, I'm just bumping to 2.15++ now until a decision can be made.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stan-dev/stan/issues/2058#issuecomment-290536631, or mute the thread https://github.com/notifications/unsubscribe-auth/AAZ_F3ojp40-rPNq2CLkYIy_8RHl4Gz_ks5rrBHmgaJpZM4J3AMe .

alashworth commented 5 years ago

Comment by seantalts Monday Apr 17, 2017 at 17:42 GMT

Is it possible this will be fixed once we upgrade to the new R / Rcpp / RTools?

alashworth commented 5 years ago

Comment by syclik Monday Apr 17, 2017 at 20:28 GMT

Yes. It's very possible. I'm not too hopeful, but if it does, then great.

On Mon, Apr 17, 2017 at 1:42 PM, seantalts notifications@github.com wrote:

Is it possible this will be fixed once we upgrade to the new R / Rcpp / RTools?

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stan-dev/stan/issues/2058#issuecomment-294538438, or mute the thread https://github.com/notifications/unsubscribe-auth/AAZ_FzsPk9P0GaA5hRYzC2B9diYsnQs_ks5rw6R5gaJpZM4J3AMe .