Hello all,
there has been a lengthy discussion at one point about how to include jupyter notebooks in repositories.
You may still be able to find it on the mailing list or in a ticket (if it indeed happened there, I am not sure).
In the end, the solution is usually to do what Erik mentioned, i. e. remove all output before committing to the repository.
We are (obviously) not the first ones hit by this and there are (more or less) convenient tools to help with this:
https://stackoverflow.com/questions/18734739/using-ipython-notebooks-under-v...
with the most feature-full solution being:
https://github.com/kynan/nbstripout
Note that all of these rely on the client to do the right thing. With bitbucket you cannot use the pre-receive hooks that git offers to reject commits that contain output (see https://jira.atlassian.com/browse/BCLOUD-10471).
Yours, Roland
Hi Erik,
Thanks for the feedback, and sorry to hear about the frustration. Indeed BaikalETK is only a proof-of-principle code at the moment, in which the second stage of development (copying the code blocks in the Jupyter notebook to a proper, callable, git-friendly Python module) hasn't been completed yet. I'm hoping that with more resources to combine ETK & NRPy+ development, the full modularization of all the ETK Jupyter notebooks can be completed.
The only way I see is to undo my changes ("git checkout"), and then pull
your changes. In this case that's fine because there are no major changes, but if I had made changes I'd be in trouble.
I generally do a `git diff` if `git pull` complains about conflicts, and it's usually easy to see if anything significant has changed (and then do `git checkout` if not). Maybe if I chose a standard way of prefixing lines that are ignorable comments (e.g., "COMMENT: Finished BSSN symbolic expressions in") a simple script could be written that checks for trivial differences like these. Removing all output isn't a good idea as it would wipe away the "standard" output displayed on nbviewer ( https://nbviewer.jupyter.org/github/zachetienne/nrpytutorial/blob/master/NRP...) in case the user wants only to learn from the Jupyter documentation.
Also if in doubt, one can simply compare the thorn output from the master Jupyter notebook with your own version. We're making it easier to redirect all NRPy+ output to whatever directory you like, so that such diffs can be made quickly and easily.
-Zach
Prof. Zachariah Etienne Physics & Astronomy Dept. West Virginia University http://astro.phys.wvu.edu/zetienne/ http://blackholesathome.net https://blackholesathome.net
On Mon, Nov 4, 2019 at 11:56 AM Erik Schnetter schnetter@cct.lsu.edu wrote:
On Mon, Nov 4, 2019 at 8:38 AM Zach Etienne < trac-noreply@einsteintoolkit.org> wrote:
#2064: Add GiRaFFE to the Einstein Toolkit Reporter: Zach Etienne Status: open Milestone: ET_2019_10 Version: development version Type: enhancement Priority: minor Component: EinsteinToolkit thorn
Comment (by Zach Etienne):
@Ian Hinder : Version control with Jupyter is only slightly more difficult than with plain source code, as its JSON is plaintext.
Is it a good idea to pull the whole code into notebooks? I don’t know how sustainable it is when you have multiple contributors; merging changes from different people can become very difficult.
Writing the Jupyter notebook is only the first step in NRPy+ development, but arguably the most important as it contains the documentation and some validation checks. I’d liken the Jupyter notebook phase more to collaboratively writing a journal article with a shared git repo than code development within such a repo.
I'd like to add a point to this, out of frustration:
A few weeks ago I downloaded and ran Tutorial-BaikalETK.ipynb. Today I try to "git pull", and it doesn't work, because there are differences in the output everywhere (e.g. "Finished BSSN symbolic expressions in 1.2987918853759766 seconds"). The only way I see is to undo my changes ("git checkout"), and then pull your changes. In this case that's fine because there are no major changes, but if I had made changes I'd be in trouble.
I think the problem is that the output is stored in the git repo. In the future, you might want to switch to a system where (a) notebooks automatically have all output removed before they are committed, and (b) upon commit, Travis or a similar system creates for-viewing-only notebooks with a standardized output.
-erik
After the notebook has been written, a separate Python module containing the code developed in the Jupyter notebook is then written. At the bottom of every NRPy+ Jupyter notebook it is required to include a self-validation check against the separate Python module (many NRPy+ Jupyter notebooks have additional validation checks). Further, all NRPy+ Python modules must include proper unit testing on all symbolic expressions generated, so that Travis CI can confirm agreement with the trusted version. This has the nice consequence of often requiring developers to update the documentation any time the module is updated in order for the Jupyter notebook to pass self-validation tests, though generally NRPy+ developers prefer updating the Jupyter notebooks as a first step, then propagating changes to the module & unit tests. Proper documentation makes our lives much easier and is the foundation of a sustainable codebase.
The above approach has worked fine with multiple contributors so far, as NRPy+ is extremely modular, with 70+ Python modules & about 100 Jupyter notebooks. Also I generally don’t put multiple students on exactly the same project (though they certainly do interact and benefit from each others' documentation). Thus
-- Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2064/add-giraffe-to-the... _______________________________________________ Trac mailing list Trac@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/trac
-- Erik Schnetter schnetter@cct.lsu.edu http://www.perimeterinstitute.ca/personal/eschnetter/
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users