Re: [Python-ideas] Re: open functions in dbm submodule need to support path-like object

I attempted to do this today, as my first actual contribution to CPython itself. I think the prior attempt went down a wrong path, which is why neither PR could actually pass tests. I've been looking at `posixmodule.c` for comparison, specifically. The key thing, I believe, is not to use `PyObject *filename` in the `dpmopen_impl()` calls at all, but rather `path_t *path`. I tried some absolutely crude instrumentation to see why the `PyOS_FSPath` macro and the rest wasn't doing what it seems like it should (i.e. sticking `printf()`s in the .c files). Eventually I realized that argument clinic is stopping me from even getting to the body of the function when `filename` was a PosixPath, so anything that PyOS_FSPath does is moot. I think in the end the new code should be even simpler than the main (3.11) branch. All I really need to do with a `path_t` is be able to call: PyObject *self = newgdbmobject(state, name, iflags, mode); That is, `name` is simply the filename as char* (the other parameters are independent of the path-like issue). Does anyone have pointers to an example or documentation about doing this? The patterns in posixmodule.c are suggestive, but I'm not sure of the missing pieces. Also, I copied the typedef for path_t from posixmodule.c to both `_dbmmodule.c` and `_gdbmmodule.c`. I know that's also going to be wrong, and it should be pulled into an .h file. Any idea about which one? `posixmodule.h` seems like a possibility, but then I'd need to repeat it in `winreparse.h`. Maybe a new name? On Tue, Sep 7, 2021 at 1:11 PM Guido van Rossum <guido@python.org> wrote:
If someone posted a Pull Request that added this, it would be looked upon favorably I'm sure.
On Tue, Sep 7, 2021 at 8:31 AM Christopher Barker <pythonchb@gmail.com> wrote:
it looks like this as had a BPO for over a year:
https://bugs.python.org/issue40563
I suggest pinging that (and maybe python-dev) to see what's up.
NOTE: first check 3.10 :-)
-CHB
On Tue, Sep 7, 2021 at 5:05 AM Evan Greenup via Python-ideas < python-ideas@python.org> wrote:
Currently, in Python 3.9, `dbm.open()`, `dbm.gnu.open()` and `dbm.ndbm.open()` doesn't support path-like object, class defined in `pathlib`.
It would be nice to add support with it.

08.09.21 08:19, David Mertz, Ph.D. пише:
I attempted to do this today, as my first actual contribution to CPython itself. I think the prior attempt went down a wrong path, which is why neither PR could actually pass tests.
I've been looking at `posixmodule.c` for comparison, specifically.
The code in posixmodule.c is a bad example, because it is too general and supports many options. It gives the patch as char* and wchat_t* (on Windows), supports file descriptors and None, and format error messages for functions supporting multiple types. But if you only need a path as char*, you can just use PyUnicode_FSConverter(). There is an existing PR for this issue. It looks correct in general, but I left some comments.

Thanks Serhiy, I've made the few additional changes you noted in the PR. I took out my attempt with path_t. However, here is why I think argument clinic (or something else?!) is actually intercepting the attempted call: With my temporary debugging, I have this function in Modules/_gdbmmodule.c: [clinic start generated code]*/ static PyObject * dbmopen_impl(PyObject *module, PyObject *filename, const char *flags, int mode) /*[clinic end generated code: output=9527750f5df90764 input=812b7d74399ceb0e]*/ { PyObject_Print(filename, stdout, 0); printf(" from _gdbmmodule.c (XXX)\n"); /* ... rest of function ...*/ And I have a very simplified test script: import _gdbm import sys from pathlib import Path print(sys.version) path = '/tmp/tmp.db' db = _gdbm.open(path, 'c') print("Opened with string path") db.close() db = _gdbm.open(Path(path), 'c') print("Opened with path-like") db.close() The output of running this is: 3.11.0a0 (heads/bpo-45133-dirty:0376feb030, Sep 8 2021, 00:39:39) [GCC 10.3.0] '/tmp/tmp.db' from _gdbmmodule.c (XXX) Opened with string path Traceback (most recent call last): File "/home/dmertz/tmp/pathlike-dbm.py", line 12, in <module> db = _gdbm.open(Path(path), 'c') ^^^^^^^^^^^^^^^^^^^^^^^^^^^ TypeError: open() argument 1 must be str, not PosixPath So before I get to the first line of the _gdbm.open() function, the TypeError is already occurring when passed a PosixPath. On Wed, Sep 8, 2021 at 3:49 AM Serhiy Storchaka <storchaka@gmail.com> wrote:
08.09.21 08:19, David Mertz, Ph.D. пише:
I attempted to do this today, as my first actual contribution to CPython itself. I think the prior attempt went down a wrong path, which is why neither PR could actually pass tests.
I've been looking at `posixmodule.c` for comparison, specifically.
The code in posixmodule.c is a bad example, because it is too general and supports many options. It gives the patch as char* and wchat_t* (on Windows), supports file descriptors and None, and format error messages for functions supporting multiple types. But if you only need a path as char*, you can just use PyUnicode_FSConverter().
There is an existing PR for this issue. It looks correct in general, but I left some comments.
_______________________________________________ Python-Dev mailing list -- python-dev@python.org To unsubscribe send an email to python-dev-leave@python.org https://mail.python.org/mailman3/lists/python-dev.python.org/ Message archived at https://mail.python.org/archives/list/python-dev@python.org/message/LQGU6DM5... Code of Conduct: http://python.org/psf/codeofconduct/
-- Keeping medicines from the bloodstreams of the sick; food from the bellies of the hungry; books from the hands of the uneducated; technology from the underdeveloped; and putting advocates of freedom in prisons. Intellectual property is to the 21st century what the slave trade was to the 16th.
participants (2)
-
David Mertz, Ph.D.
-
Serhiy Storchaka