NEO/go doesn't understand what it should do with these parameters and
raises an error in case they are added to the NEO URI [1].

[1] https://lab.nexedi.com/kirr/neo/-/blob/f1a1bb9d/go/neo/client.go#L551-553

/reviewed-by @kirr @jerome
/reviewed-on !1620

'wendelin.core' needs to differentiate between a NEO cluster with SSL
and without SSL from a normalized zurl (e.g. a zurl which doesn't
contain any paths to encryption files). Therefore the 'neo' / 'neos'
scheme difference was introduced in nexedi/neoppod@bc3e38ea.
This patch introduces this difference in zurl that SlapOS provides for
'wendelin.core' - this essentially fixes a ERP5+WCFS setup that connects
to a NEO cluster with activated SSL.

/reviewed-by @kirr @jerome
/reviewed-on nexedi/slapos!1620

We only need one query identifier ('?'), while query options should be
separated by the separator identifier ('&'). This essentially changes
the resulting NEO URI from

  neo://cluster@IP&a=b?b=c

to

  neo://cluster@IP?a=b&b=c

/reviewed-by @kirr @jerome
/reviewed-on !1620

This reverts commit 0cf70a6e.

NEOgo and NEOpy diverged wrt the NEO URI format. Due to NEOgo
changes in the NEO URI format 0cf70a6e was necessary.
However in neoppod!18 and
neoppod!21 a common
URI format for NEOgo and NEOpy was agreed on. Therefore most changes of
0cf70a6e are no
longer necessary. We still need the difference between 'neo' (without
SSL) and 'neos' (with SSL), but we'll re-introduce this in the next
commit.

This revert keeps the changes made in 6fa1d9ae as this is still
needed.

/reviewed-by @kirr @jerome
/reviewed-on !1620

New 'wendelin.core' version contains several bugfixes that were
discovered in field usage. Furthermore it introduces the new 'ZBlk_fmt'
option 'auto' that heuristically sets ZBlkfmt to gain a good ratio between
read time and database size, improves python3 compatibility and synchronizes
NEO zurl scheme between NEO/py and NEO/go.

See nexedi/wendelin.core@63153845...db6fea3d
for complete list of changes.

/reviewed-by @kirr @jerome
/reviewed-on nexedi/slapos!1620

don't test it in python2 anymore

it seems the RPC deleteToken never returned NOT_FOUND anyway.

we are not using this command line utility, in erp5 tests (
product/ERP5Type/tests/Python3StyleTest.py ) we are using the python
API directly, not the command line.

if developer needs to invoke 2to3 it is present in ${python:location}/bin/

Also, md5sum option has been renamed runtime_plc_md5sum.

This is a fixup of 7df1d5fe.

this does not work when serving the raw blob on gitlab, so it is not
always usable for parameter editor.

The ISO used is a modified version of the Online version (no
repositories bundled in the CD) with updates
(SLE-15-SP6-Online-x86_64-QU1-Media1.iso) available in
https://www.suse.com/download/sles .

The only modification done is to set the "timeout" variable in
/boot/x86_64/loader/isolinux.cfg to 0, so that the KVM does not
automatically boot from disk after a timeout if the user does not press
a key (which was a problem that we found while testing this distro that
made difficult to install it).

See merge request !1655

And revert "software/slapos-testing: Quickfix python -E errors" now
that pygolang's pymain supports -E option.

This reverts commit 0fad9a6d.

Use md5sum to check if what we download is what we actually got.

GTK tried to compile against libcups when it was present in the system.
This failed the compilation if the includes were not present.

This commit sets "file" as the only backend for printing, thus
explicitly removing cups as a dependency.

See merge request nexedi/slapos!1649

Always install cups in lib folder, even when lib64 would be used by
default.

See merge request !1649

The multiarch strings (GNU target triplets) used currently for platform
detection are different in Suse, as they include the "vendor" ("suse" or
"pc" depending on the GCC version) and the OS part is just "linux"
instead of "linux-gnu".

This changes platform detection to use the constants available in
buildout for detecting the operating system and the `platform` module
for detecting the machine, instead of these multiarch strings.

See merge request !1648

follow up 917296b4 (pylint 1.4: Understands six.{PY2,PY3} import
guards (already supported on pylint 2)., 2024-09-25)

Several components use patches directly from Github.
Unfortunately, Github changes the number of characters displayed for the
commits in index lines depending on the size of the repository.
That means, that the content of these patches may change, invalidating
the MD5.

By using the `?full_index=1` param
(https://github.com/orgs/community/discussions/46034#discussioncomment-4846112)
we can request that these patches include the full names of the commits,
thus preventing future changes.

See merge request !1645

if libnsl was installed on the system, openvpn would link with it
so be proactive and always link with slapos libnsl.

This patch from PR 147 wa a left over from development, when we merged
we used AccessControl 7.0 which included this patch.

 * https://github.com/mroonga/mroonga/commit/f4d3e6a612d346f690bbf68893e339ff0b69e5b8 might fix the crash we were observing with https://lab.nexedi.com/nexedi/slapos/-/blob/0fad9a6dcaf6f5750c03dcd2a6b42364207767ff/software/erp5/test/test/benchmarks.py , I ran this test 10 times without crash.

 * ~~There's a test failure in ERP5, but it does not seem harmful and we can have a workaround ( erp5!1993 )~~ In the end, the problem was fixed upstream so we backport the fix.

See merge request !1644