[package] conan-boost: cant install the boost::mpi library

[rasmus98]

I dont know if this is a bug per say, but it is at least a problem which makes a part of the boost library uninstallable, and is easilly fixable. I would create a pull request if I knew how to, sorry...

When boost is installing, the boost::mpi library along with assosiated components in various other boost libraries are only istalled if the following line is added to the user-config.cam:

using mpi ;

This file is being autogenerated by the conanfile.py, so it is not possible to just add this line manually. An easy fix would be to add the following code on line 748 of the conanfile.py:

        if self.options.custom_userconfig:
            contents += '\n' + self.options.custom_userconfig

And we could thus in our conanfile.txt specify,

[options]
boost::custom_userconfig="using mpi ;"

or similar, I dont quite understand how string literalts work in conan.txt files

[SSE4]

boost MPI requires an actual MPI implementation, such as OpenMPI, Microsoft MPI or Intel MPI. they need to be packaged first.

[rasmus98]

Yes yes absolutely, in order to compile it! Of cause it would be nice if conan could also handle this part, but for most places where this library would be used (clusters, supercomputers) the system admins have already installed (multiple) MPI implementations. I think most users of boost.mpi would be well aware that this is something they would have to install themselves externally to conan (like the rest of the compiler).

The point is just that without this change, I dont know that i can use conan to install boost at all, since it it just missing this one build flag which i cant add.

[SSE4]

what we can do is to add some options to specify mpi implementation, then conan will include that during the generation of user-config.yml.

[oleurodecision]

Hi,

Can't the recipe just rely on the without_mpi recipe option ? Something like

        if not self.options.without_mpi:
            # https://www.boost.org/doc/libs/1_74_0/doc/html/mpi/getting_started.html
            contents += '\nusing mpi ;'

Later, optional settings could also be set up,

using mpi
   : [<MPI compiler wrapper>]
   : [<compilation and link options>]
   : [<mpi runner>] ;

as most (all ?) of these values can be retrieve from the mpi wrapper (using --showme variants, see mpi wrapper manpage for details)

(Note that I'm not an advanced MPI user, only a consumer facing the same problem here)

[SSE4]

@oleurodecision something like that could work, yes.
however, that needs an error handling for the following case:

1. user specifies without_mpi=False, he wants to build with MPI support
2. there is no valid MPI installation on the machine
3. boost silently turns MPI off, build successfully, but without MPI libraries
4. as result users gets package with incorrect package id - it's labeled as built with MPI support, but actually it wasn't

this needs to be tested if there is an error handling to be added to prevent such incorrect builds.
if you can confirm that it's not needed (e.g. you tested it, and boost refuses to build instead off silently turning MPI support off), then I think we're good to go for the first iteration.
feel free to submit PR in such a case.

[oleurodecision]

see #2842

seems there is things to fix in config.yml/conandata.yml KB-H052 (not related to current PR though)

[NashTools]

@oleurodecision
i added the changes you proposed in #2842 but it still does not work for me. i get ./boost/mpi/config.hpp:20:10: fatal error: mpi.h: No such file or directory during compilation. However, the logs say the MPI Auto-configuration was successful. I have openmpi installed using apt-get in ubuntu.`

[oleurodecision]

okay thanks for the feedback, I will try to look at this soon.

…

________________________________ De : NashTools <[email protected]> Envoyé : samedi 19 décembre 2020 00:20 À : conan-io/conan-center-index <[email protected]> Cc : Olivier LIESS <[email protected]>; Mention <[email protected]> Objet : Re: [conan-io/conan-center-index] [package] conan-boost: cant install the boost::mpi library (#2501) @oleurodecision<https://github.com/oleurodecision> i added the changes you proposed in #2842<#2842> but it still does not work for me. i get ./boost/mpi/config.hpp:20:10: fatal error: mpi.h: No such file or directory during compilation. However, the logs say the MPI Auto-configuration was successful. I have openmpi installed using apt-get in ubuntu.` — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#2501 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AMC34CH5EEKWTNMUTYPQIMLSVPPSFANCNFSM4P5MGCMA>.

[oleurodecision]

Just tested on ubuntu 18.04 server "vanilla" (fresh install in vm), using apt install mpi-default-dev for mpi setup, everything seems ok, see attached files.

boost-1.72.0-default.log
boost-1.72.0-mpi-no_headers.log
boost-1.72.0-mpi.log
conan-boost-packages.txt

[oleurodecision]

Note that #2842 was closed and replaced with #3959 (needed to change branch name 😞 )

[NashTools]

@oleurodecision first of all thank you for your reply
i checked out #3959 and simply executed conan create . boost/1.72.0@ -o boost:without_mpi=False in the folder /conan-center-index/recipes/boost/all
here are the logs
output.txt
on line 15450 and subsequent logs you can see mpi.h was not found
Im using gcc-10 though. Any hint is very much appreciated.

[oleurodecision]

Can you indicate which ubuntu version you are using please ?
I'll try to test tonight, else we'll need to wait a few days (first holidays for months 🎉)

[NashTools]

im using ubuntu 20.10 groovy gorilla

[oleurodecision]

Tested with success :

• on ubuntu 10.04
  • using gcc-7
• on ubuntu 20.04
  • using gcc-9
  • using gcc-10
• on ubuntu 20.10
  • using gcc-10

let's check that more precisely in a few days ;)

[NashTools]

i finally got it running. The issue was on my side, somehow openmpi was not installed correctly. To all who has the same problem look up here https://stackoverflow.com/questions/60982859/why-openmpi-package-from-apt-is-incomplete?noredirect=1&lq=1

Again thank you @oleurodecision for your support

[oleurodecision]

you're welcome (btw, url markdown seems wrong)

[perseoGI]

Hi there!
I'm closing this issue as it seems to be already addressed.
If problem persists, do not hesitate to reopen this issue or create a new ticket.
Happy coding 🐸