Skip to content

Latest commit

 

History

History
285 lines (183 loc) · 9.24 KB

CONTRIBUTING.md

File metadata and controls

285 lines (183 loc) · 9.24 KB

Dependencies

Building NEMO requires following minimum dependencies:

  • Unix tools, such as: make, csh, bash, git
  • C/C++/Fortran compiler
  • pgplot (optional, but highly recommended)

git

There are two ways you could have obtained the NEMO source code:

  1. You cloned from the upstream : https://github.com/teuben/nemo . This is fine if you just want to compile and run, but not ideal if you make modifications and want to share them back to the upstream via a pull request (PR). You can however "repair" your local repo, discussed below, and still submit a PR.

  2. You forked NEMO from the upstream, and cloned it locally from the repo in your own github account. This is the ideal method, but you will still need to set the upstream manually if you used github.com. See also the gh command below for an even easier way.

  3. Sadly on github.com you will also find a zip copy of the repo that does actually work fine, except it's a frozen snapshot and cannot be efficiently used to collaborate. However, if you cannot install git, this is probably the only way to bootstrap yourself. For example https://github.com/teuben/nemo/archive/refs/heads/master.zip, which will create a directory nemo-master. Other branches are available through similar zip file construct.

Familiarize yourself with the concept of a pull request on github. There are some links at the bottom of this document.

gh: github CLI

You can safely skip this section if you prefer to work via github.com, though the gh command described here is by far the fastest and easiest way to work with the github ecosystem. You just have to intall yet another tool for this.

If you can use conda, installation can be done as follows:

  conda install gh --channel conda-forge

but see also manual installing instructions, after this you need to authenticate once via your github account:

  gh auth login

after which you could create your own fork, clone locally, and set the upstream all in just one step:

  gh repo fork https://github.com/teuben/nemo

If all is well, the following commands should show the correct origin and upstream:

  git remote -v

  origin [email protected]:YOURNAME/nemo.git
  upstream [email protected]:teuben/nemo.git

for both the (fetch) and (push). None of these should be blank! You are now ready for working in your own branches and issue a pull request (PR).

2. Cloning your personal fork

This section can be skipped if you use the gh repo fork method described before this.

If you have cloned your own fork on github.com, you should now clone it

  git clone https://github.com/YOURNAME/nemo

you only need to set the upstream:

  git remote add upstream https://github.com/teuben/nemo

and you are ready for creating a PR (from a branch of course).

1. Cloning the official upstream

If you happened to have cloned the official upstream

  git clone https://github.com/teuben/nemo

then things are a bit more complicated, because you should have cloned your fork. However, the following commands will fix this (assuming you also went to github.com and made that fork):

  git remote rename origin upstream
  git remote add    origin https://github.com/YOURNAME/nemo

Again, the gh command now gives a single line shortcut to all this:

  gh repo fork https://github.com/teuben/nemo

Keeping your clone in sync

You should regularly make sure your local master branch is in sync with the upstream master branch. This allows you to work in local branches, and be up to date by branching off the tip of this upstream master branch. Here's a recipe for that in your local clone:

  git checkout master
  git fetch upstream
  git merge upstream/master
  git status
  git push

Working in a branch

Assuming your own master is in sync with the upstream master, here is a typical example to work in a branch, using a branchname b1

  git branch b1
  git checkout b1
  
  ## edit/add your files; test them; commit them, e.g.
  git commit -m "my comment"  existing_file
  git add new_file
  git commit -m "my command"  new_file
      
  git push -u origin b1

Now you can issue a pull request on this branch b1.

If you have become a fan on the gh pr method, here's the recipe for this:

  git checkout -b b1
    <<edit, test, commit>>
  gh pr create

once the branch has been merged, you don't need it locally anymore, so delete it

  git branch -d b1
  git push origin --delete b1

Common work in a branch

Lets say there is a new development idea, lets call it "table2", and others will share the development in this "table2" branch, but until the development is done, this is not merged to the master yet.

  1. NEMO (upstream) itself will start making this new development branch:

    git checkout -b table2 ... git push --set-upstream origin table2

  2. others will do then branch off this new branch (user astroumd is used as example here):

    gh repo fork https://github.com/teuben/nemo nemo cd nemo git remote add upstream https://github.com/teuben/nemo git fetch upstream git checkout -b myTable3 git merge upstream/table2 ... git push --set-upstream origin myTable3

    on github.com/astroumd/nemo you can then do a pull request from astroumd::myTable3 to teuben:table2. An alternative (next item) is that the upstream person would pull in myTable3 and tests locally

  3. NEMO, optionally, creates an alias site "astroumd" and merge from their myTable3 branch to check out the code

    git checkout table2 git remote add astroumd https://github.com/astroumd/nemo git pull astroumd git merge [--no-ff] astroumd/myTable3 .... (resolve conflicts) git push

  4. after this, all collaborators will need to merge these back:

    git checkout myTable3 git fetch upstream git merge upstream/table2

Memorable git options

  1. Show all files modified in a branch AAA

    git diff master...AAA --name-status --diff-filter=M

  2. When was a branch created

    git show --summary git merge-base AAA master gitk --all --select-commit=git merge-base foo master

  3. To see which files belonged in a commit, find the sha (e.g. via "git log" or "git log file" if you know the file that belonged to it), then

    git diff-tree --no-commit-id --name-only -r SHA

  4. Difference between two SHA's

    git diff

PR with github CLI

Recapping working with "github CLI" (gh)

Step 1 from the submitter of the PR:

  gh repo fork https://github.com/teuben/nemo
  cd nemo
  git checkout -b teuben1
  $EDITOR index.md
  gh pr create

Once the branch has been merged by the upstream, there is no need to keep it. It can be removed as follows:

  git branch -d teuben1
  git push origin --delete teuben1

Step 2 by the receiver of the PR:

Tests

From the top level directory in NEMO there are a few basics regression tests and benchmarks:

  make check
  make bench5
  make bench

The check target depends on the many Testfile files sprinkled throughout NEMO.

The bench5 target currently runs 6 NEMO programs all designed to run 5 seconds on some particular processor that your NEMO compilation can now be compared to. The default speed for that processor is defined to be 1000, anything larger is a faster processor. Currently in 2023 we're approaching a single core score of 2000!

The bench target is still under development, and has a more diverse set of programs, all controlled by sprinkled Benchfile files. See the script src/scripts/nemo.bench.

Debugging

Most applications are built using make variables defined in the $NEMOLIB/makedefs file. Although not recommended, hacking is always possible by directly editing this file, with the caveat you are then bypassing the configure step of the install, which also does modify some other files. Notably the falcON package has some extra dependancies that start with configure.

NEMO uses a set of hierarchical Makefiles, and many things (see google) have been written why they are bad. But we don't recurse!

Debuggging NEMO applications

Consult the local Makefile of the application. You should be able to override the compiler (CC=, CXX=, FC=) and related options. If the application was integrated into NEMO, there should be a line

  include $(NEMOLIB)/makedefs

in that Makefile, and check the Makefile which are used.

An example:

  cd $NEMO/usr/trenti/CGS
  make clean install FC=flang FFLAGS=-O3
  make bench2

would create an alternate (and as it happens twice as fast as with gfortran) of the CGS N-body integrator.

References

Some references on git workflows: