Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Oct 02 04:31
    github-actions[bot] labeled #2776
  • Oct 02 04:31
    github-actions[bot] commented #2776
  • Oct 02 04:31
    github-actions[bot] labeled #2771
  • Oct 02 04:31
    github-actions[bot] commented #2771
  • Oct 01 04:39
    github-actions[bot] labeled #2770
  • Oct 01 04:39
    github-actions[bot] commented #2770
  • Oct 01 04:39
    github-actions[bot] labeled #2767
  • Oct 01 04:39
    github-actions[bot] commented #2767
  • Oct 01 04:38
    github-actions[bot] closed #3706
  • Oct 01 04:38
    github-actions[bot] labeled #3706
  • Oct 01 04:38
    github-actions[bot] closed #2407
  • Oct 01 04:38
    github-actions[bot] labeled #2407
  • Sep 30 20:17
    conda-bot commented #4588
  • Sep 30 20:17
    ScottBailey synchronize #4588
  • Sep 30 18:28
    conda-bot commented #4588
  • Sep 30 18:28
    ScottBailey reopened #4588
  • Sep 30 18:28
    ScottBailey closed #4588
  • Sep 30 18:23
    conda-bot commented #4588
  • Sep 30 18:23
    ScottBailey synchronize #4588
  • Sep 30 16:27
    conda-bot commented #4588
Marius van Niekerk
@mariusvniekerk
@msarahan you might be able to do it with Mergify
Marcelo Duarte Trevisani
@marcelotrevisani

@msarahan you might be able to do it with Mergify

That is interesting, I was taking a look on that and it seems to be simple to set and very useful

Matt
@mattevans_gitlab
Is there a way to force channeldata.json to be regenerated when performing a channel index?
Matt
@mattevans_gitlab
I ask because I cannot for the life of me get a package's description to be updated in channeldata.json. It's updated everywhere else
I haven't rm -rf'd the entire dir yet (noarch), but would like to avoid that if possible. I'm going to debug into the index now but it seems odd to me that so much is involved to have it update outdated information
Matt
@mattevans_gitlab
ok well just deleting the channeldata.json/index.html and re-indexing forced it to update. Unless I'm missing something that's weird. In this case the version didn't change, but the description of the package did (and the tar.bz2 package was overwritten in the channel directory)
Isuru Fernando
@isuruf
Can someone from anaconda team look at AnacondaRecipes/r-base-feedstock#1 ?
Isuru Fernando
@isuruf

@mingwandroid, do you know why there are two copies of binutils in the package?
For ex:

x86_64-conda_cos6-linux-gnu/bin/as
bin/x86_64-conda_cos6-linux-gnu-as

above two files are the same file (checked with sha256sum), but have two different inode numbers

Ray Donnelly
@mingwandroid
Really we should use a symlink here. But AFAIR GCC calls read link on the exe which would follow the symlink, then GCC might read the filename to determine the triplet etc. I know clang does that determination at least. (OK you are talking about binutils instead but it could be similar)!
Need to get back into compiler stuff soon. Got python cve things to look at for now though.
Isuru Fernando
@isuruf
Ah, they are hardlinks on the build environment, but conda-build doesn't care about it I guess
matrixbot
@matrixbot

gabm > <@gitter_mingwandroid:matrix.org> Need to get back into compiler stuff soon. Got python cve things to look at for now though.

on this note: are there gdb builds available as well? my system gdb doesn't seem to work very well with the assemblies of the conda gcc

Paul Madden
@pmspire
I am building the NetCDF Fortran library as part of a larger recipe and, with --error-overlinking enabled for conda-build, I get an ERROR ... runpaths ['$ORIGIN/.'] found in ... libnetcdff.so.6.2.1 failure. Is this really a problem? Isn't this what conda-build would patchelf it with anyway? Is there something I can do to appease the rpaths checker?
Paul Madden
@pmspire
I was able to overcome this by running patchelf --remove-rpath on the affected executables (I'll try the same for libraries) before they're examined by conda-build. Checking the final executables in the conda .bz2 archive, it does look like conda-build writes back an identical rpath, if I'm not mistaken.
Ray Donnelly
@mingwandroid
@matrixbot I believe it wouldn't be too bad to update our compilers so that GDB is provided, yes, both for build=>host (with gdbserver for host and gdb for build) since we use crosstool-ng on Linux. Maybe when I get back to compiler stuff I can try to schedule time to look into that in more detail. Can you list the features you need though when building GDB please? Would it extend to Python integration? That could be trick for me mentally to wrap my head around and would create a ecosystem-bootstrapping problem for me too .. unless I were to do something awfully hacky
Isuru Fernando
@isuruf
Is there a way to tell conda-build that I know what I'm doing
conda_build.exceptions.CondaBuildException: Empty package; python dep present in build but not host requirements.
Nehal J Wani
@nehaljwani
@isuruf Does adding a version constraint on it help?
Isuru Fernando
@isuruf
What do you mean?
Tim Snyder
@timsnyder
@mingwandroid I'm interested in having gdb as well. Currently, I just build it as a separate package that depends on the compilers but I'm not using it for actual cross tools, just x86 on x86.
Ray Donnelly
@mingwandroid
@timsnyder OK, thanks for the feedback. We've got some compiler work coming up I think, so I will definitely try to get as much gdb as I can going. BTW, is python integration essential for you? I'm kind of thinking it is these days and may have to break my own promise never to create a bootstrapping loop in the ecosystem for this :-(
Ray Donnelly
@mingwandroid
@isuruf I am happy to hardcode in conda-build, if check_output('id -nu') == 'isuruf': context.user_knows_what_user_is_doing = True ;-)
Not sure it'll pass review though!
Isuru Fernando
@isuruf
Haha. I found a workaround anyway
Duncan Macleod
@duncanmmacleod
hi all, I'm trying to use compiler('m2w64_c') to compile on windows, but the compiler needs to be used through python. When i run conda build it is still using my local cl.exe, is there a trick to get it to actually use the m2w64 gcc?
Duncan Macleod
@duncanmmacleod
nm, I think I found a good example here
Duncan Macleod
@duncanmmacleod

that got me a good bit further, but I eventually got down to

ld.exe: cannot find -lmsvcr140

is this something other people have seen, and is there a way around it?

Duncan Macleod
@duncanmmacleod
this looks like an issue specific to my windows box, which doesn't show up in azure devops, sorry for the noise
Isuru Fernando
@isuruf

Anyone seen an error like,

Fixing permissions
'\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' not in tarball

https://ci.appveyor.com/project/conda-forge/islpy-feedstock/builds/27506201/job/79wnggh4ymdv5skg

Elmar Pruesse
@epruesse
Does anyone have an idea how to work around the #!....placeholder/bin/perl problem with max #! line length being 127?
I have a package where I need to use the perl version from the conda env (to build the right binary perl module).
Just seding to /usr/bin/env perl doesn't work because of /usr/bin/env perl -w and because I want the path to be rewritten later.
(Some of the scripts are used during build /and/ distributed)
Jonathan J. Helmus
@jjhelmus
@isuruf Most of conda team and Anaconda has been looking into that issue for the last two days
Isuru Fernando
@isuruf
Thanks @jjhelmus
Jonathan J. Helmus
@jjhelmus
something is adding \x00 to the "info/files" file
Isuru Fernando
@isuruf
it doesn't happen in azure though
Probably because cph 1.4.1 is used in azure and 1.5.0 in appveyor
Jonathan J. Helmus
@jjhelmus
yes it seems to have something to do with the interplay between cph and conda-build
Isuru Fernando
@isuruf
Thanks @jjhelmus for the info. Is this a consequence of static libraries?
Jonathan J. Helmus
@jjhelmus
That was the one of our initial thoughts but Ray was able to reproduce the bug using an older cph prior to the static library
Ray Donnelly
@mingwandroid
well, to be exact mine is likely slightly different and more to do with cph create specifically .. Until https://github.com/mingwandroid/conda-package-handling/commits/wchar/src/conda_package_handling I was unable to use a filelist that contained utf-8 filenames as an input to cph create and I really needed to for testing. Still, our wchar stuff is broken but the info/files thing I think you've got a better handle on than I @jjhelmus.
I'm not quite tired yet, so I am happy to jump back in for an hour or so if you can point me at any part of this?
Now, even with my wchar branch, I can make a .tar.bz2 with unicode filenames correctly, but only via tarfile and not via libarchive. Still interesting that you report success with 1.4.1. I wonder why that is?
Jonathan J. Helmus
@jjhelmus
There is something up with \r\n being converted to \n and then appending nulls somewhere in libarchive. I'm trying to track it down
Ray Donnelly
@mingwandroid
ok. I'm hacking about with the split recipe, and adding patching from git as an option to try to make it a bit easier to hack on (for me!)
Jonathan J. Helmus
@jjhelmus
sounds good
Ray Donnelly
@mingwandroid
going for e.g.:
{% set dev_nature = os.environ.get('CONDA_BUILD_DEBUG_PACKAGE', '') %}
{% if dev_nature != '' %}
  {% set dev_nature = "_dev" %}
  {% set patches = "libarchive-patches.git" %}
{% else %}
  {% set patches = "libarchive-patches" %}
{% endif %}
then git can be used and I'll fetch from git if you set CONDA_BUILD_DEBUG_PACKAGE (maybe more fine grained)