Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • 06:21
    allanleal commented #891
  • 04:19
    hmaarrfk commented #871
  • 04:14
    hmaarrfk commented #871
  • 03:42
    hadim commented #871
  • 03:39
    hmaarrfk commented #871
  • 03:18
    hadim commented #871
  • 03:05
    hmaarrfk commented #871
  • 01:51
    isuruf commented #871
  • 01:43
    isuruf commented #871
  • 01:42
    hadim commented #871
  • 01:41
    hadim commented #871
  • Nov 12 23:47
    beckermr commented #891
  • Nov 12 23:47
    beckermr commented #891
  • Nov 12 21:06
    hadim commented #871
  • Nov 12 20:26
    isuruf commented #921
  • Nov 12 20:24
    wasimaftab commented #921
  • Nov 12 20:24
    wasimaftab commented #921
  • Nov 12 20:24
    wasimaftab commented #921
  • Nov 12 20:14
    isuruf closed #921
  • Nov 12 20:14
    isuruf commented #921
Isuru Fernando
@isuruf
@mingwandroid, if you want to, you can create an activation script and move the file to RECIPE_DIR/PreLoad.cmake. This is an undocumented feature in cmake though
Ray Donnelly
@mingwandroid
OK, I may check in then with the CMake team to make sure they don't plan to remove it!
Ray Donnelly
@mingwandroid
Great, nice. You've been busy!
If we can avoid a load of CMake checks we could really speed things up. One preload per macOS SDK for example or Linux CDT distro .. all the common stuff! (really I find people who complain about autotools being slow to be generally unfair if they claim CMake faster, in my experience they remain comparable relative to project size)
(we can also do this with a config.site for autotools if anyone wanted to persue that! I'm great at coming up with ideas for others, I'm sorry)
Isuru Fernando
@isuruf
Haha
Ray Donnelly
@mingwandroid
Adding these things would move the pointer towards being able to offer faster builds for our packagers. Yay.
and you know, faster builds for us
then I want optional (when it works, when the testsuite is executed to completion) parallel distutils/setuptools/pip (does pip have any such support yet?)
I wish Brad would comptemplate breaking an undocumented feature in the interests of making it a legit feature!
Usually suggesting hiding such a change behind a "policy" flies with Brad, but I guess this is lower level than policies or read in earlier so that's not possible. Still better semantics could be introduced behind a differently named file. Seems simple enough to me'
(apart from you know, all that coding of those new semantics and whether to complicate the old code or add a new C++ file for it)
Isuru Fernando
@isuruf
I have a longer list of long term TODOs.
  • Break out sysroot and openmp
  • Switch from gomp to llvm-openmp
  • Add clang_linux-64 for linux
  • Add clang_win-64 for linux
  • Support cmake's bindexplib on linux
Ray Donnelly
@mingwandroid
What is bindexplib? Is it a symbol versioner thing? Or a tdb/importlib type thing?
Isuru Fernando
@isuruf
It creates .def files from object files
Ray Donnelly
@mingwandroid
I would add: facebook's BOLT to our conda-build backend as a very worthwhile thing to explore, and upx support.
Isuru Fernando
@isuruf
that's what CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS uses
I want that for autotools
Ray Donnelly
@mingwandroid
for these things I'd lean on LIEF to be honest with you.
we already call for starters
and I don't intend to back that out at present, it can be slow, so I'm making sure to call it once per sha1 of binary file only (will cache things when done and settled too, somewhere). LIEF's doing good IMHO as a project.
need to find out more about your plans here though
Matthew R. Becker
@beckermr
Idk how to fix it and it is really causing problems for me
Ray Donnelly
@mingwandroid
@beckermr our team is completely preocuppied at present, please, if you can dig into it yourself
Matthew R. Becker
@beckermr
sounds good. I’m preoccupied too of course!
Uwe L. Korn
@xhochy
Lately I've seen a lot of PRs where noarch: python was removed because conda-build complained about an exe on Windows that looked like an entrypoint, e.g. here conda-forge/staged-recipes#9588 (build logs don't resolve correctly anymore for me).
Anyone has an idea if this is an undeclared entrypoint or what is happening here?
Chris Burr
@chrisburr
Uwe L. Korn
@xhochy
Thanks!
Duncan Macleod
@duncanmmacleod

@duncanmmacleod - you need to run VsDevCmd.bat -arch=amd64 to set up the correct environment variables for conda build to see the msvc compiler

thanks @dhirschfeld, I tried this and got some new path entries, but still no cl.exe, if I run VsDevCmd.bat -test that fails because it too cannot find cl.exe. I presume I've done something wrong in installing VS in the first place, will investigate

Filipe
@ocefpaf
Folks, what would be the best pattern/practice to make proj.4 (old name) 'conflicts with' proj (new name)?
Chris Burr
@chrisburr
@ocefpaf see here
Duncan Macleod
@duncanmmacleod
@dhirschfeld, I installed an extra set of VS2019 components, and now have VsDevCmd.bat working fine, and conda-build can now see the compiler! thanks for your help!
Filipe
@ocefpaf
Thanks @chrisburr!
Wolf Vollprecht
@wolfv
hey all, i am trying to make ROS's distribution model fit in conda-forge. this is what we came up with so far:
there is a ROS mutex pacakge that carries the distribution string (e.g. melodic)
and there is a ros-conda-base package that run_constrains some base dependencies (they are specified in this "REP" : https://www.ros.org/reps/rep-0003.html )
maybe someone cares to review this? getting this one merged will be a huge step towards opening the floodgates for the rest of ROS :)
the catkin package is the ROS meta build and workspace tool in the ros ecosystem and therefore basically the most important component
Jaime Rodríguez-Guerra
@jaimergp
Hey! I am seeing some weird differences between Appveyor and Azure for Windows. My package contains some py files with hardcoded prefix paths that should get rewritten by conda-build. In Azure, these are caught, but for some reason they are ignored in Appveyor. The only difference is that Azure uses C:\Miniconda\conda-bld and Appveyor goes for C:\bld... I can reproduce the Appveyor behavior in my windows box, which also uses C:\bld. Is that the reason? The PR is here conda-forge/staged-recipes#9163
Jaime Rodríguez-Guerra
@jaimergp
Well, that was not the reason, but ripgrep! If rg is present, the file is not found.
and the hardcoded prefix gets ignored
Matthew R. Becker
@beckermr
Can someone on core pull conda versions 4.7.11 and 4.7.12? Both of those versions break staged-recipes.
Sylvain Corlay
@SylvainCorlay

conda versions 4.7.11 and 4.7.12 [...] break staged-recipes.

:unamused:

Matthew R. Becker
@beckermr
software is hard :(