-
-
Notifications
You must be signed in to change notification settings - Fork 4.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
WIP: python #84
WIP: python #84
Conversation
Should be an interesting one. I expect conda-build-all to work nicely in this repo, but suspect that we will trip up when it comes to conda-building the recipe in the feedstock (as a special environment var needs setting |
Looks like we should have the unxz package on the docker image to decompress xy files. |
I've been getting a similar failure locally. Looks like readline isn't building correctly. |
|
Python.org also supplies a .tgz file which might be a better choice. The manylinux1 Docker image does not include xz (and unxz) but it is available from yum. |
There are some different between this recipe and the python-3.5 recipe available at ContinuumIO/anaconda-recipes. I am able to build the latter with only minor modification using the manylinux Docker image. |
Yeah, I have seen that readline issue before when building Python. Will need to check my notes to see what the solution was. |
@@ -0,0 +1,56 @@ | |||
package: | |||
name: python | |||
version: "3.5.1" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should we do a series of these packages and name them with major minor version?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, could extract all of these version numbers and use jinja to make this simpler to maintain.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No, I'd like the feedstock to called <package-name>-feedstock
. My aspiration it to use branches to manage different versions (as you would maintenance branches in the source repo).
Support is very loose at the moment in conda-forge, but the first repo to do this already exists (https://github.com/conda-forge/gdal-feedstock) and it is only a matter of time until I need to deal with it 😉
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sounds reasonable. Is there some discussion of this happening somewhere? It would be nice to iron out how this works before we get a bunch of things like gcc
, libgcc
, etc.
The lack of effort on this PR (on my behalf), is a representation of my inbox over the last few days. I'm keen to move it forwards, but in truth, there is more value in me supporting of of the other super contributors. Essentially, I'm pausing this until I get some more time. |
Sorry about my part in that. Just really excited about leveraging this infrastructure. 😄
Thanks for supporting us.
Sounds reasonable. I was just interested after seeing @jjhelmus work on this, as well. It would definitely be great to get a fully operational Python into our stack here. |
There are recipes for Python 2.7, 3.4 and 3.5 at ContinuumIO/anaconda-recipes which seem promising and I've gotten the 3.5 recipe working on Linux with some minor modification. I am not sure what exactly the One question that should be discussed is if we want to "brand" the Python like Continuum or this PR does. Personally, I do not think we should, we don't do to any other package so why is Python special. |
Definitely seems like the right place to start.
I think I agree with this. Just to play devil's advocate though, one reason we might want to brand is to make it clear that they have are not using Continuum's brand of Python and they are not using the vanilla system Python. |
Yep. That is the reason IMO. |
Whatever the decision is on branding, I think we should carry it forward with |
Also, another good place to look for Python and things like this is |
I have 3 reasons (of different quality):
Whilst I'm completely against obtrusive / unconstructive branding, in this case I think it is helpful to the conda-forge community to have it. |
Depends on the branding. 😄 Also, I don't think just because we can do it that we should.
That's true. Though it is never difficult to check (i.e.
This is also why if we do this it should be done with things like
Agreed. Though I think this is almost an argument against branding. Without the branding file, they can just drop in their own branding file. If they want to use this one, they need to change what they are doing to fit this (so changes in
👍
Again I feel like this is more of an argument against branding then for it basically for the same reasons mentioned above. |
This is a new one on me:
|
Is running |
I have seen this before. Took a moment to track down. Try setting the macro |
Normally, I have seen this issue with CMake though and it needs another option. |
Not the only one: openmm/openmm#1419 |
Ping @mingwandroid - he might know more about this. On Wed, Mar 23, 2016, 04:09 Phil Elson notifications@github.com wrote:
|
This is the CMake issue's solution. Maybe that will get us going forward down the right path at least. |
Seems the install_name_tool pre-installed on travis-ci is also broken with respect to -delete_rpath. This was something I added to conda-build to workaround SIP for OS X 10.11 which prevents DYLD_FALLBACK_LIBRARY_PATH from doing anything (DYLD_* in fact since Apple have classified them as exploit vectors). To see the original issue, remove the following from conda-build/conda_build/environ.py:
.. and then the following from conda-build/conda_build/macho.py:
and then try to rebuild r-base on 10.11. You should find it fail to load conftest due to not being able to load gfortran,3.dylib. I propose, for now, to add some hacks to detect the broken versions of install_name_tool and not do -delete_rpath when they are detected, then to initiate a parallel effort to rebuild the old binaries with fixes for this bug and to encourage travis-ci (and anyone else who is known to run them) to adopt the fixed ones (provided everything I've said here is correct and can be verified). Does this sound reasonable? |
I'd be interested to know what version you get from:
.. they didn't make that easy, unless I missed something obvious :-) |
Proposed reversion for now: conda/conda-build#844 |
One thought here might be to actually switch to the latest version of Mac and XCode/clang here and actually just compile in compatibility mode with 10.7. This might also solve the OpenMP problem by just having support in the system compiler. Though I don't know how this would be affected by the compatibility mode. |
Hi! This is the friendly conda-forge-admin automated user. I wanted to let you know that I linted all conda-recipes in your PR ( Here's what I've got:
|
@jjhelmus You're correct. the |
Superseded by #645. 🎉 |
No description provided.