Python works with GNU DBM and with Berkley DBM.
Berkley DBM has been under Oracle for some time.
And it's not clear how many Python users actually use DBM.
In the packages feed, we have both libdb47 (which is now under Oracle) and
GNU DBM. The GNU DBM has a compatibility layer for Berkley DBM.
There are newer versions than libdb47, but it's probably not worth having
them yet. The libbd47 tarball is ~40+ MB. Odds are newer versions will be
bigger and more bloated.
This change merges the old `python3-gdbm` package into the `python3-dbm`
package, since they are effectively using the same underlying library now,
i.e. gdbm.
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
I seem to forget to check/select setuptools and pip (that come bundled with
Python).
This change will do a simple 'ls' on the 2 wheel files, so that the build
fails even if just building Python.
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
Dropped patches:
0004-Replace-EVP_CIPHER_CTX_cleanup-with-EVP_CIPHER_CTX_r.patch
0005-Switch-get_-Update-APIs-to-get0.patch
Reworked patches:
0001-Add-new-ASN1_STRING_get0_data-API.patch
0006-Add-X509_STORE_CTX_trusted_stack-compatibility-macro.patch
These 2 require that we keep only the CUSTOMIZATIONS stuff for now. Maybe
later we can drop this.
Ran 'make package/python-cryptography/refresh'.
Added patch:
0004-disable-rust.patch
upstream did a sloppy job with the CRYPTOGRAPHY_DONT_BUILD_RUST logic; we
need to patch it, to make sure the setuptools-rust isn't installed.
We may need to carry this patch in our tree for a bit longer than upstream,
because in newer versions, CRYPTOGRAPHY_DONT_BUILD_RUST logic gets removed.
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
By adding these dependencies, it is much easier for users (both
applications that use Twisted and end users) to have secure
communications by default.
Signed-off-by: Jeffery To <jeffery.to@gmail.com>
Python3 comes with a built-in readline module. It wasn't included up until
now; mostly because it wasn't considered.
This change introduces it as a sub-package of the main Python3 package.
readline support is included in Python.
libreadline pulls libncursesw as a package, so python3-ncurses was
updated to pull libncursesw as well.
It should be the same package; mostly done for consistency.
Resolves the issue reported here:
https://forum.openwrt.org/t/python3-repl-missing-readline/90039
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
Refreshed Python patches.
Updated pip & setuptools version.
For pip, patch '001-pep517-pyc-fix.patch' was reworked.
Also, the current version of the bundled pip (21.1.1) no longer supports
Python2, so the 'py2.py3' suffix gets replaced with just py3.
For setuptools, there is no longer a script/module:
https://github.com/pypa/setuptools/pull/2544
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
Related to discussion:
https://github.com/openwrt/packages/pull/14060
Every once in a while a version bump will occur that requires an ABI
change. Example: Python 3.8 to 3.9. When this happens some Python packages
would need to be rebuilt. In setups where everything gets rebuilt, this
isn't a problem.
It's usually a bigger problem when needing to upgrade something via
opkg.
To accommodate for this, we add a libpython with it's own ABI_VERSION
flag. If this ABI_VERSION changes, then this should propagate forward.
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
Fixes: https://github.com/openwrt/packages/issues/15370
This is inspired from:
330bc94dcc/lang/python-greenlet/Makefile
The `PKG_USE_MIPS16:=0` is not taken into consideration when building
Python modules. That's because the sysconfig is used.
This is only an issue with greenlet (on MIPS) so far.
One option is to do `PKG_USE_MIPS16:=0` in the core Python package.
But, since we know that the `wlanslovenija` group has successfully used
greenlet on MIPS with this construct, we might as well adopt it until GCC10
becomes the main compiler.
As noted here:
https://github.com/openwrt/packages/issues/15370#issuecomment-817015484
GCC10 doesn't have this problem.
Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
Alexandru Ardelean
Jeffery To
Jan Pavlinec
Javier Marcet
Daniel Golle
Josef Schlehofer
Martin Matějek