Reworked the neuron.hclass3 module
#1096
Conversation
Helveg
changed the title
hclass3 moduleneuron.hclass3 module
Codecov Report
@@ Coverage Diff @@
## master #1096 +/- ##
==========================================
+ Coverage 31.42% 31.47% +0.04%
==========================================
Files 571 572 +1
Lines 108767 108863 +96
==========================================
+ Hits 34182 34261 +79
- Misses 74585 74602 +17
Continue to review full report at Codecov.
|
|
Regular tests go in |
Helveg
force-pushed
the
hclass-nometa-nonlocal
branch
from
f8b5966
to
23e3064
Compare
Helveg
force-pushed
the
hclass-nometa-nonlocal
branch
from
fab6723
to
20a6e95
Compare
|
thanks @Helveg! 👌 (this is in draft. once ready for review, tag Michael and Robert for the review) |
share/lib/python/neuron/hclass3.py
Outdated
| if len(hoc_bases) > 1: | ||
| bases = ", ".join(b.__name__) | ||
| cname = cls.__name__ | ||
| raise TypeError(f"Composition of {bases} HocObjects not allowed in {cname}") |
There was a problem hiding this comment.
Note: f-strings require Python 3.6+
There was a problem hiding this comment.
I see... Python 3.5 goes through a different file.
There was a problem hiding this comment.
Yea, Python 3.5 or earlier just keeps using the old hclass3.py file since the required improvements to metaclasses were simply not available in earlier versions, so I kept things the same there.
|
In order for |
|
I added tests and make some minor changes such as allowing composition if the HOC type is the same and requiring |
|
Is this becoming ready? In that case, we can |
|
It's ready, there's one open action point by @ramcdougal to integrate the benefits of |
|
There are some CI failures but Github has changed some default CI images yesterday/today. And that's why we have been seeing some random failures. Hopefully this will be fixed in coming days. @ramcdougal : this PR looks good for you as well? |
share/lib/python/Makefile.am
Outdated
| @@ -85,4 +86,3 @@ EXTRA_DIST = $(nobase_neuronhomepython_DATA) | |||
| # if test $(NRNPYTHON_PYMAJOR) -gt 2 ; then \ | |||
| # $(PY2TO3) -w $(DESTDIR)$(neuronhomepythondir) ; \ | |||
| # fi | |||
|
|
|||
There was a problem hiding this comment.
nitpick: avoid trivial changes
share/lib/python/neuron/Makefile.am
Outdated
| @@ -1,5 +1,3 @@ | |||
| ## Files which get installed in $(prefix)/share: | |||
|
|
|||
| SUBDIRS = rxd crxd | |||
|
|
|||
There was a problem hiding this comment.
nitpick: avoid trivial changes... in this file especially, there's no need to change this file at all
|
@Helveg : would love to get this PR ready with 8.0 release. @ramcdougal added comments but I see that those are now marked as resolved as well. Anything else is pending here? 🚀 |
|
There's still some things to be done, I'll get to that right away! |
|
🎉 thank you @Helveg ! @alexsavulescu : please cherry-pick this for 8.0! |
* added pyenv's .python-version to gitignore * prevent import py2/3 module in py3/2. removed `exec` wrapper from hclass * Changed comment and removed non-existing `nonlocal_hclass` import * Removed traling whitespaces and double newline at eof * Removed the MetaHocObject metaclass and added `nonlocal_hclass` factory * New `HocBaseObject`; (nonlocal_)hclass uses this new base. * Added dev note to demystify quintessential link between HOC & Python * moved __all__ to PEP8 location * Fixed error with grandchildren of `HocBaseObject` * Apply changes only to Python 3.6+ where metaclass alternative is availbl * fixed import typo * hclass of hoc_type should precede use of hoc.HocObject API's * Fixed usage of module_name * added hclass35.py to Makefile, excessive trailing newlines. * `module` does not exist, see details on mixed approaches in PR#1096 * black formatting & updated hclass2 derived docstrings * Store dummy modules on `neuron` module to better emulate submodules * Add a TypeError if init but not new has been given (see #1129) * Added _pyobj_enabled flag to check whether Python object interface is up * Explicitly defined __new__ for 3.6+ pyobj compatibility * Class composition is allowed if HOC types are the same * added tests for HOC type Python classes, test only in Python 3.6+ * Removed 'super()' call for Python 2 * Revert "added hclass35.py to Makefile, excessive trailing newlines." This reverts commit 080f14a. * Added hclass35 to Makefile * Revert "Removed traling whitespaces and double newline at eof" This reverts commit 046fab4. * Reinstated deprecation messages * Merged hclass and nonlocal_hclass
* added pyenv's .python-version to gitignore * prevent import py2/3 module in py3/2. removed `exec` wrapper from hclass * Changed comment and removed non-existing `nonlocal_hclass` import * Removed traling whitespaces and double newline at eof * Removed the MetaHocObject metaclass and added `nonlocal_hclass` factory * New `HocBaseObject`; (nonlocal_)hclass uses this new base. * Added dev note to demystify quintessential link between HOC & Python * moved __all__ to PEP8 location * Fixed error with grandchildren of `HocBaseObject` * Apply changes only to Python 3.6+ where metaclass alternative is availbl * fixed import typo * hclass of hoc_type should precede use of hoc.HocObject API's * Fixed usage of module_name * added hclass35.py to Makefile, excessive trailing newlines. * `module` does not exist, see details on mixed approaches in PR#1096 * black formatting & updated hclass2 derived docstrings * Store dummy modules on `neuron` module to better emulate submodules * Add a TypeError if init but not new has been given (see #1129) * Added _pyobj_enabled flag to check whether Python object interface is up * Explicitly defined __new__ for 3.6+ pyobj compatibility * Class composition is allowed if HOC types are the same * added tests for HOC type Python classes, test only in Python 3.6+ * Removed 'super()' call for Python 2 * Revert "added hclass35.py to Makefile, excessive trailing newlines." This reverts commit 080f14a. * Added hclass35 to Makefile * Revert "Removed traling whitespaces and double newline at eof" This reverts commit 046fab4. * Reinstated deprecation messages * Merged hclass and nonlocal_hclass
I rewrote the
hclass3module as it is the most important link and entry-point into thehocmodule for object oriented Python developers that want to understand where their HOC objects are coming from. In its original form this was quite hard to figure out and it was using some outdated/problematic idioms (see #1092, #1094, #1093) mostly only of importance once people start actually creating and inspecting subclasses of the Python interface objects.I unpacked the strings in
hclass2andhclass3(closes #1092). I added a factory class that fills in the required class module attributes for inspection (closes #1094). I switched out the localhcclass for a better nonlocalHocBaseObjectthat does not require a metaclass (closes #1093) and added a PEP8 compliant__all__attribute. I also think,hclass3.pyshould be renamed to_hclass3.pyas it is an internal interface, but that will be part of another, larger move toward a PEP compliantneuronpackage in general ;).(For clarity and PEP compliance I did rename the
.htypeattribute to._hoc_type, but I think it's only used to pass along the hoc type to the__new__function anyway)I added RST docstrings to all the function and added a dev comment for anyone that would like to know how their HOC object types are generated. I also made sure that all invalid use of the functions resulted in clear exceptions raised where they occurred.