[FIX] immunize shutil.rmtree to node non-existence for remove_node_di… #3135
Conversation
…rectories=True in the case that stop_on_first_crash=False
Codecov Report
@@ Coverage Diff @@
## master #3135 +/- ##
=========================================
Coverage ? 67.59%
=========================================
Files ? 299
Lines ? 39480
Branches ? 5220
=========================================
Hits ? 26685
Misses ? 12082
Partials ? 713
Continue to review full report at Codecov.
|
The ``InterfaceHelpWriter`` class was practically a clone of ``ApiDocWriter``. Both have been merged into one single module, having the Interface helper inherit from the ApiDocWriter.
This PR follows up on nipy#3119 (merge that one first for a clean diff, or diff against ``oesteban:maint/dedup-apigen-code``). In practice, this PR fixes several broken points of our documentation (e.g., the workflows list was empty and now it has been updated, changelog not rendered, API of pure python code not rendered by the Nipype API parser was missing, etc.). CHANGES ------- * Replaced the ``numpydoc`` sphinx extension with ``sphinxcontrib-napoleon``. * Removed autosummary sphinx extension, required by numpydoc * Cleared up ``docs/sphinxext/*``, as nothing is now used from there * Use current sphinx-apidoc/autodoc/autosummary * Removed the modref generation tooling, as it is not necessary anymore after re-enabling apidoc. * Cut building warnings down to 321 - just those we incur because our API generator. This required some fixes of some docstrings. Beyond those corresponding to the Nipype API generator, only missing links remain as warnings (for sections in the navbar). * Updated changelogs to be reStructuredText.
Building on top of nipy#3119 and nipy#3129, this PR makes a deep revision of the documentation: * Added a new ``build_docs`` job to CircleCI to test how it renders. * Minimized external machinery (under ``/tools/``) when building the documentation: 1. Some minimal modifications to sphinx extensions (apidoc, napoleon) allow the generation of special documentation for nipype interfaces, as it used to be before this PR 2. A new sphinx extension (``nipype.sphinxext.apidoc``) takes care of parsing and rendering inputs and outputs. They now look like the parameters/arguments of functions when formatted with numpydoc. * Revised the description of many interfaces and the documentation of the main class and the input/output specs. * Revised the structure of the navbar, separating out User-Guide/Examples, Interfaces-Index, and Devs' documentation. * Minimized the number of WARNINGS at documentation build to 5 (4 of them coming out from the auto-generated SEM tools).
Co-Authored-By: Chris Markiewicz <[email protected]>
Co-Authored-By: Chris Markiewicz <[email protected]>
Co-Authored-By: Chris Markiewicz <[email protected]>
Co-Authored-By: Chris Markiewicz <[email protected]>
Co-Authored-By: Chris Markiewicz <[email protected]>
In particular, after allowing printing the inheritance of interfaces, links for ZZZCommandBase interfaces were broken, as originally filtered out and not built. Now they are being built and we are back to 5 warnings.
By removing some space concatenation of strings, some of them were r-strings and the concatenated one contained ``\n``, effectively escaping the special return-carriage. Instead of concatenating strings, the interface now accumulates the lines in a list that is joined in the end.
Master is broken -- this PR relocates the pip install of ``niflow-nipype1-workflows`` so that it happens AFTER nipype was installed.
|
Do you have a test? |
There was a problem hiding this comment.
A regression test would be nice. If infeasible, I'm comfortable merging, as ignore_errors has been around for all supported versions of Python.
This should be rebased onto maint/1.4.x, and it would probably be best if we wait on #3146 to rebase.
|
I could write a regression test for this, but it might be overkill? I can't really think of a case where ignore_errors=True would cause issues. As you noted, the flag has been around since Python 2. I do like the idea of waiting to rebase this into #3146 though, rather than rush the merge. |
|
The regression test would be to make sure future changes don't restore this problem, not whether this fix causes new problems. |
Omnibus backport for 1.4.x
Ahh, I see what you mean now. |
…rectories=True in the case that stop_on_first_crash=False
…rectories=True in the case that stop_on_first_crash=False
|
@dPys This history looks a bit complicated. I would suggest the following to clean it up: git fetch upstream
git checkout -b fix/rmtree_ignore_errors upstream/maint/1.4.x
git cherry-pick 1c98d4d
git push -u origin fix/rmtree_ignore_errorsThen create a new PR, closing this one. |
|
Replaced by #3148. |
Summary
Acknowledgment