2020-07-01 conda-forge core meeting
Attendees
Agenda
Your agenda items
-
(WV) Quick announce of micromamba (https://gist.github.com/wolfv/fe1ea521979973ab1d016d95a589dcde)
-
(WV) Update on standardization of next gen package format from last meeting?
-
(MRB/IF) cos7 and CDTs plans (https://github.com/conda-forge/conda-forge.github.io/issues/1085)
-
cos7 compilers are working!
-
we have removed any mention of cos6 or cos7 from the sysroot path so that we have one compiler
- old sysroot:
x86_64-conda_cos6-linux-gnu
- new sysroot:
x86_64-conda-linux-gnu
- old sysroot:
-
we now use the sysroot/glibc/kernel headers packages directly from cos6 or cos7 when building (instead of rebuilding glibc)
-
this has created an issue where CDTs get installed to the wrong path (sym links kind of work, but one would not be able to swap cos6 to cos7 in an env)
-
thus we want to rebuild the CDTs to remove cos6 and cos7 from their paths (and will use the sysroot packages to make sure old-style CDTs, cos6 and cos7 are all mutually exclusive in envs)
-
old-style CDT:
- needs to have
no_hoist
in the source sections - sysroot dir has
conda_cos6
orconda_cos7
in the path - needs to have
run_constrained
entry ofsysroot_{subdir} ==99999999999
to prevent it from being co-installed w/ the new compilers or CDTs
- needs to have
-
new-style CDT:
- needs to have
no_hoist
in the source sections - sysroot dir has
conda
only in the path - needs to have
run
requirement on the proper version of thesysroot_{subdir}
package so that it is only installed with CDTs from the right version of CentOS- 2.12 for cos6
- 2.17 for cos7
- needs to have
-
old plan is to roughly
- patch repo data to make current CDTs on conda-forge conflict w/ new compilers (PR issued)
- move all of the needed CDTs from defaults to conda-forge so that we can build from conda-forge (in progress)
- turn on strict channel priority (or even remove defaults from list of channels for linux) for builds
- write a migrator that will issue PRs to change all old-style CDTs to new-style CDTs
- remove path shims from new compilers and sysroots
-
which ones to move etc
- list of needed things here: https://github.com/beckermr/misc/blob/master/work/conda_forge_cdt_scripts/needed_cdt_list.yaml
- keep
- selinux, mesa, pciutils etc (IDK what is included in etc here)
- X11
- ca-cert is a dep of the java CDTs
- libxt (use by gnuplot and r-base)
- maybe keep
- alsa ones (used by qt but we have this packaged too: https://github.com/conda-forge/alsa-lib-feedstock)
- libpng-devel (needed for wxpython and javafx-sdk)
- numactl (used by a lot of feedstocks)
- not keep
- zip (but not in list also)
-
new plan!
- put run_constrained in defaults and conda-forge
- build special repo to build them and upload them outside of feedstocks (both new and old kinds)
- build and upload the minimal set to conda-forge
- turn on strict channel priority in builds and/or remove defaults from channels
- push final builds of the new compilers
-
-
(FF) Should we do Outreachy as part of an effort to support diversity in tech? Advantages are low cost and high impact. Dissdvantages are the time effort from the mentors.
-
(UK) CFEP-18: (Not) Packaging static libraries
-
(IF) cf-mark-broken: Marking not broken packages as broken
- marking packages broken breaks envs
- repo data patches are always better
- we might want to allow maintainers to mark things broken when the consumers of them are only themselves
- the speedy aspect of doing things is good for pushing bug fixes fast
- to do (MRB)
- 24 hour time limit
- add more detail to PR text to help ppl understand what they are doing
- make it easier to revert
-
(KK) CUDA 11 support
- CUDA 11 dropped CentOS 6 support
- ties into CentOS 7 migration above
cudatoolkit
11 - https://github.com/AnacondaRecipes/cudatoolkit-feedstock/pull/7- Make sure we have a glibc run constrain line
- Add Docker image (need to double check how
FROM
can be adjusted conditionally)- Needs a recent docker version but this should work
ARG IMAGE=ubuntu:lastest
FROM ${IMAGE}
- Needs a recent docker version but this should work
-
(CJ) Anyone want to put in for PyData Global?
- https://global.pydata.org/pages/cfp.html
- (submissions not open yet)
- CJ will put something in under "Lessons from Industry" discussing a risk model view of CF (please let CJ know if you are interested/have a perspective to add)
-
(IF) strict channel priority
- Needed for PyPy
- Needed for Cos7
- Remove packages that we don't fully support. For eg: there are 3 builds of scipy on windows.
Active votes
Subteam updates
Bot
- (CJ) Status of openjdk, r-java stalled out
Stuff from last week that we didnt get to
ARM
POWER
CUDA
Docs
staged-recipes
website
security+systems
- still need to finish CFEP-13