This section contains notes for O₂scl developers.
Make sure version numbers are updated in
and the tables in
doc/o2scl/sphinx/download.rstare updated with “not yet set” for the new version hashes.
For most recent commit, make sure the tests, examples, and documentation all succeed. Check source installs on ubuntu and OS X and a homebrew HEAD install.
Try a test build on docker
Make sure snaps are working using
snapcraft build --debug.
Update NEWS file with recent changes.
Make the final commit targeted for release.
Check the commit succeeds on docker
Promote the snaps on snapcraft from edge to beta at https://snapcraft.io/o2scl/releases .
Refresh the documentation using
Create the new distribution and copy to internal svn repo.
Create github release, tagging the recent commit and uploading the distribution.
Compute hashes with
openssl dgst -sha256and update dl_page.dox with hashes and github release hash. Regenerate documentation with
make installon isospin so that
make utk-sync-doccan copy docs from the post-installation directory
Copy the new distribution and the new sha256 hash to https://isospin.roam.utk.edu/public_data/o2scl_dists/
Update homebrew recipe with the new version number and new hash.
Check installation using homebrew directly.
Turn on build pushes in travis-ci.org and commit again since
Procedure for moving to new development version¶
Update to the new development version number and new OLIB numbers in
Update version numbers in doxyfile files and main.dox files and in dl_page.dox .
Update local configure scripts to refer to new version number if necessary
o2sclpy release procedure¶
Update version numbers in
Regenerate the o2sclpy documentation using
make docand upload it to web using
Remove old dists in o2sclpy by clearing o2sclpy/dist directory
python3 setup.py sdist bdist_wheel
Upload a new version of o2sclpy to pypi using
twine upload dist/*
Coding recommendations and guidelines¶
Comment code liberally.
Avoid goto statements.
Avoid hard-coded numerical values which are not either: (i) well-commented, or (ii) parameters modifiable by a class-user.
Match the existing coding style when possible.
Make sure all lines fit in 78 columns.
Because this is a library (and not a end-user executable), it is generally better to call the error handler when an input is incorrectly specified instead of assuming some alternative correct value.
All new classes deserve new documentation and new testing code.
Documentation and code should be in standard American English.
All code must be released in GPLv3.
Global variables and static member data should be avoided.
When reasonable, put input parameters first and output parameters last.
When possible, templated vector parameters with size_t arguments should appear similar to
size_t &n, vec_t &v, and in that order.
All code should be ANSI-compatible, and, inasmuch as is possible, operating system and platform independent.
Exception messages should be as informative as possible.
Exception messages should end with the full class and function name from where they are thrown.
Functions which deallocate memory should never fail and should never be required to call the error handler. Also, class destructors should never be required to call the error handler.
Avoid passing pointers. Pass either bare objects (which then require copy constructors), shared pointers, or const references.
For numeric parameters to functions, if not all values are permissible, then the error handler should be called when a non-permissible value is given by the user and the function documentation should clearly explain why.
Wherever possible, objects should be usable by default. Avoid zombie objects which are instantiated but unusable. An important exception to this rule is in classes which would otherwise have to load data from a file in order to operate normally, as file I/O should not occur in constructors (so that MPI can instantiate classes without worrying about parallel I/O).
Objects should thread-safe in the weak sense, that is, different processes should be able to safely access and modify different instances of the same class at any time. Functions which read (but not modify) class data should be thread-safe in the strong sense, that is, different processes should be able to read the same instance of a class at any time.
Whereever possible, ensure your code compiles without warnings using flags analogous to the gcc string:
-ansi -pedantic -Wno-long-long -Wall -Wno-unused -Wextra -Wconversion -Wshadow -Wpointer-arith -Wcast-align -Wwrite-strings
Avoid ‘try’ blocks, as a goal is that o2 should compile with -fno-exceptions.
Functions which return c void should end with
All functions which are called by the constructor should be documented as doing so
Object destructors should almost never call the error handler.
All functions which have more than one input reference of the same type should be clear if they allow multiple references to the same object
Refer to other classes with \ref if necessary. Refer to function parameters with \c or embed them in html TT (text-type) commands.
Bibliographic references should be used. When possible, include the DOI link which begins with the prefix http://dx.doi.org (not the vendor-specific DOI link).
Comment Doxygen documentation with \comment and \endcomment. (Yes, sometimes comments in comments are useful.)
Communicate with the lead developer before, during, and after any non-trivial development. Communicate your ideas before development, so that you don’t write many lines of code only to find that your pull request will be rejected. Communicate your ideas during development to avoid conflicting changes. Communicate your ideas after development to ensure they have a chance of being implmented. Subversion is not a replacement for real communication.
Pull requests will be integrated into the trunk by the lead developer at whatever time they deem appropriate.
Developer-specific files which are not platform-independent should not be added to the repository. Sometimes
.gitignorecan be used to ignore these files, but this should be done sparingly.