adding IPOPT recipe #1607

pstjohn · 2016-09-16T15:21:55Z

Borrows heavily from https://github.com/SCIP-Interfaces/conda_recipes/tree/master/ipoptlib, this library likely deserves to have its own conda-forge recipe. Currently compiling against MKL on OSX and Linux.

Will depend on

mumps_seq Add recipe for mumps #1445
asl adding recipe for ampl-mp #1778

conda-forge-linter · 2016-09-16T15:21:58Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/ipopt) and found it was in an excellent condition.

tkelman · 2016-09-18T13:47:31Z

recipes/ipopt/build.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+THIRDPARTY=(Metis Mumps)


Be aware that the version of Metis downloaded here has a strange license.

You'll need ./get.ASL too if you want an executable.

Thanks for the heads up! Do you have a suggestion for a better place to get Metis? There is a conda feedstock, but it only builds a static library. Its also a different version (5.1.0 vs 4.0.3 used from get.Metis)

Just added ASL in the most recent commit, hope that works.

Metis is optional in Ipopt. You'd need to use Mumps 5.x for it to work with Metis 5.x (which has a normal Apache license), and I don't think Ipopt has been updated yet to work with Mumps 5.x. Maybe if you build them all separately it might work as an external library instead of building the vendored copy under ThirdParty, if there aren't API changes. I think homebrew-science might be building Mumps and Ipopt that way.

Probably simplest to leave out Metis for now if you want to redistribute binaries (kind of the point of conda-forge), and revisit with a todo for building latest versions of things separately. If you're going to link against MKL then you should be able to use the MKL implementation of Pardiso for Ipopt anyway, which won't need Metis.

conda-forge-linter · 2016-09-18T17:16:48Z

Hi! This is the friendly automated conda-forge-linting service.

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

conda-forge-linter · 2016-09-18T20:49:45Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/ipopt) and found it was in an excellent condition.

conda-forge-linter · 2016-09-18T21:04:20Z

Hi! This is the friendly automated conda-forge-linting service.

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

conda-forge-linter · 2016-09-18T21:39:13Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/ipopt) and found it was in an excellent condition.

conda-forge-linter · 2016-09-18T22:05:41Z

Hi! This is the friendly automated conda-forge-linting service.

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

conda-forge-linter · 2016-09-19T02:19:37Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/ipopt) and found it was in an excellent condition.

conda-forge-linter · 2016-09-19T02:52:48Z

Hi! This is the friendly automated conda-forge-linting service.

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

conda-forge-linter · 2016-09-19T13:48:38Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/ipopt) and found it was in an excellent condition.

basnijholt · 2016-10-18T20:05:18Z

@Jvanschoubroeck this will ease your pain 😄

pstjohn · 2016-10-18T20:14:59Z

Glad to see there's interest! But this might still be a while before we merge. We'd rather not build third-party libs in each recipe, see coin-or/COIN-OR-OptimizationSuite#3.

There's also the question of whether or not we're allowed to compile against intel's MKL from #1445

basnijholt · 2016-10-18T20:28:05Z

I would just go with openblas for now.

The MKL provided by anaconda does something to your code that it automatically parallelizes single core operations, which is rather inefficient and what you don't want when you parallelize your own code.

AFAIK you can't compile code without a MKL license, but when the code is already compiled (by someone with a license) you are free to spread it.

Disclaimer: I am far from an expert on any of these subjects.

tkelman · 2016-10-18T21:01:32Z

It's not just BLAS and LAPACK that's the issue here, Ipopt can also use the Pardiso sparse linear solver in MKL which is much faster than the only redistributable option MUMPS.

pstjohn · 2016-10-19T00:37:00Z

But you can load a separate linear solver after the fact using shared libraries. I wonder then how much the blas/lapack you supply to configure really matter in that case.

tkelman · 2016-10-19T01:08:49Z

Have you used the linear solver loader with MKL's Pardiso? I know it works with HSL solvers and with the non-MKL version of Pardiso but I've never heard of anyone using it with a dynamically loaded MKL.

tkelman · 2016-10-19T01:10:47Z

Usually Ipopt is dealing with sparse problems and BLAS only gets called on small dense sub blocks so in my experience the BLAS implementation makes much less difference than the sparse solver. It'll depend on specific problem structure though.

tkelman · 2016-10-19T01:47:40Z

Looking over https://projects.coin-or.org/Ipopt/ticket/216 where we originally implemented linking Ipopt against MKL's Pardiso, since the APIs differ between MKL and non MKL Pardiso and the switch in Ipopt is made based on a compile time flag, I doubt the linear solver loader will work. Maybe if you swap the defaults...

pstjohn · 2016-10-19T02:45:55Z

Gotcha, thanks for weighing in! You'd obviously know this better than me, I've never actually used IPOPT with Pardiso.

Where does that leave this recipe? Are you familiar with how COIN currently ships binary versions for IPOPT?

tkelman · 2016-10-19T03:02:17Z

As a big monolithic installer that has all of coin-or, or via distributions that have packaged the individual components. All the openly redistributable binaries use MUMPS as far as I know. I used to maintain the Ipopt matlab interface which was a mex file, and was built in a special way to use the copy of HSL MA57 that comes with matlab. Ampl and Gams also ship their own executable or library that they build, not positive which linear solvers they include. In JuliaOpt we leverage homebrew on mac and have cross compiled binaries for windows that we download from the opensuse build service.

What's the specific question? There are lots of ways to build this for different requirements.

pstjohn · 2016-10-19T11:35:20Z

Cool, thanks for the background. The question was what would be the most useful to users looking for IPOPT: it seems that if someone wants to use the MKL pardiso solver, they're going to have to compile from source anyways (vs. grabbing a binary from COIN or elsewhere).

So would a conda install that links against MUMPS & openblas be the best implementation we're allowed to distribute?

pstjohn · 2016-12-22T21:05:19Z

CI builds are passing! 🎉
RTM on my end, let me know if I'm missing anything

ocefpaf · 2016-12-22T21:08:05Z

recipes/ipopt/meta.yaml

@@ -0,0 +1,58 @@
+# Note: there are many handy hints in comments in this example -- remove them when you've finalized your recipe
+
+# Jinja variables help maintain the recipe as you'll update the version only here.


Please remove the example comments from your final recipe.

ocefpaf · 2016-12-22T21:09:03Z

recipes/ipopt/meta.yaml

+
+test:
+  files:
+    - Cpp_example


I would name this directory test but that is OK.

moorepants · 2016-12-29T22:19:25Z

recipes/ipopt/meta.yaml

+    - scotch
+    - gcc
+    - blas 1.1 {{ variant }}
+    - openblas 0.2.18|0.2.18.*


This is now at 0.2.19 I believe. Might as well update now.

moorepants · 2016-12-29T22:19:50Z

recipes/ipopt/meta.yaml

+
+extra:
+  recipe-maintainers:
+    - Peter St. John


I think this should be your github username.

basnijholt · 2017-01-20T10:28:11Z

Could this be merged? :)

pstjohn · 2017-01-20T14:25:31Z

I'm done with it, do you know who I should ping to get it merged? @conda-forge-admin ?

jakirkham · 2017-01-20T14:28:12Z

recipes/ipopt/LICENSE

@@ -0,0 +1,87 @@
+Eclipse Public License - v 1.0


This doesn't need to be in the recipe as license_file is able to get this from the source.

jakirkham · 2017-01-20T14:28:56Z

recipes/ipopt/meta.yaml

+    - mumps
+    - metis
+    - scotch
+    - gcc


Do we need the whole compiler at run time or just libgcc?

I'm assuming we need the whole compiler, mainly because the test compiles a test program and makes sure the solution is optimal. I could cut these though if that's not warranted.

g++ -I$PREFIX/include/coin -c -o cpp_example.o cpp_example.cpp ... ./cpp_example | grep -q "Optimal Solution"

Normally one adds gcc to test/requires then.

Done below.

Note: Diff comments have moved from run to build.

jakirkham · 2017-01-20T14:57:18Z

recipes/ipopt/meta.yaml

+
+about:
+  home: https://projects.coin-or.org/Ipopt/
+  license: EPL


Would add 1.0 as it appears to be that version of EPL.

jakirkham · 2017-01-20T14:58:35Z

recipes/ipopt/meta.yaml

+about:
+  home: https://projects.coin-or.org/Ipopt/
+  license: EPL
+  license_file: LICENSE


Seems the license file at this level is just filler.

For licensing information for files in this directory, please see the comments in the header of each file. For licensing of files in each of the subdirectories, please see the LICENSE file in that subdirectory.

Instead would use license_file: Ipopt/LICENSE, which appears to be the actual EPL 1.0 license.

jakirkham · 2017-01-20T16:17:22Z

Thanks @pstjohn. Will merge when it passes.

pstjohn · 2017-01-20T16:24:17Z

Thanks for the help finishing this up!

tkelman reviewed Sep 18, 2016

View reviewed changes

pstjohn changed the title ~~adding IPOPT recipe~~ [WIP] adding IPOPT recipe Sep 18, 2016

pstjohn mentioned this pull request Sep 18, 2016

Conda-forge recipe casadi/casadi#1868

Closed

pstjohn force-pushed the ipopt branch 2 times, most recently from 798f15e to 3b4064c Compare September 19, 2016 02:19

pstjohn force-pushed the ipopt branch from a734725 to e30b730 Compare September 19, 2016 02:52

pstjohn mentioned this pull request Sep 20, 2016

Recipes for conda-forge coin-or/COIN-OR-OptimizationSuite#3

Closed

pstjohn force-pushed the ipopt branch from 88b2ad7 to b73f429 Compare September 20, 2016 22:43

This was referenced Oct 17, 2016

adding recipe for ampl-mp #1778

Merged

Add recipe for mumps #1445

Merged

pstjohn added 9 commits December 22, 2016 09:29

adding epl license file

1d0f819

moving to pkgsource

5d201a8

fixing asl recipe name :(

10fb6e8

addressing review [skip appveyor]

b0549f5

adding mumps_mpi.h patch

778b759

building with gcc, adding gfortran linker flag

9b30521

adding metis/scotch dependencies

d654b44

adding asl include dir [skip appveyor]

7749bff

testing gcc_eh include [skip appveyor]

f7849da

pstjohn changed the title ~~[WIP] adding IPOPT recipe~~ adding IPOPT recipe Dec 22, 2016

ocefpaf approved these changes Dec 22, 2016

View reviewed changes

addressing review [skip appveyor]

031ff53

moorepants reviewed Dec 29, 2016

View reviewed changes

recipes/ipopt/meta.yaml

extra:

recipe-maintainers:

- Peter St. John

Copy link

Contributor

moorepants Dec 29, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this should be your github username.

pstjohn added 2 commits January 9, 2017 08:32

addressing @moorepants review

23576a8

correcting openblas test version

757bde9

jakirkham reviewed Jan 20, 2017

View reviewed changes

removing license file

3be4f17

jakirkham reviewed Jan 20, 2017

View reviewed changes

pstjohn added 2 commits January 20, 2017 08:31

more refinements

af8ba60

resetting license location

ae9c776

jakirkham approved these changes Jan 20, 2017

View reviewed changes

jakirkham merged commit 450ffa8 into conda-forge:master Jan 22, 2017

		@@ -0,0 +1,58 @@
		# Note: there are many handy hints in comments in this example -- remove them when you've finalized your recipe

		# Jinja variables help maintain the recipe as you'll update the version only here.

adding IPOPT recipe #1607

adding IPOPT recipe #1607

Conversation

pstjohn commented Sep 16, 2016 • edited

conda-forge-linter commented Sep 16, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

conda-forge-linter commented Sep 18, 2016

conda-forge-linter commented Sep 18, 2016

conda-forge-linter commented Sep 18, 2016

conda-forge-linter commented Sep 18, 2016

conda-forge-linter commented Sep 18, 2016

conda-forge-linter commented Sep 19, 2016

conda-forge-linter commented Sep 19, 2016

conda-forge-linter commented Sep 19, 2016

basnijholt commented Oct 18, 2016

pstjohn commented Oct 18, 2016 • edited

basnijholt commented Oct 18, 2016

tkelman commented Oct 18, 2016

pstjohn commented Oct 19, 2016

tkelman commented Oct 19, 2016

tkelman commented Oct 19, 2016

tkelman commented Oct 19, 2016

pstjohn commented Oct 19, 2016

tkelman commented Oct 19, 2016

pstjohn commented Oct 19, 2016

pstjohn commented Dec 22, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

basnijholt commented Jan 20, 2017

pstjohn commented Jan 20, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jakirkham commented Jan 20, 2017

pstjohn commented Jan 20, 2017

pstjohn commented Sep 16, 2016 •

edited

pstjohn commented Oct 18, 2016 •

edited