Like for af0c5b11

    $ git diff 8.5.1+ce.0-1-ge732b39..8.6.5+ce.0-0-g342f8be --
        files/gitlab-cookbooks/gitlab/templates/default/sv-sidekiq-run.erb

is empty.

Like for 61177775

I have manually reviewed:

git diff 8-7-stable 8-8-stable -- files/gitlab-config-template/gitlab.rb.template \
                                  files/gitlab-cookbooks/gitlab/attributes/default.rb

on omnibus-gilab repository.

There are only changes on postgresql and registry options, which are not relevant for us.

NOTE: 8-7-stable is 8.7.9+ce.1-0-gf589ad7
      8-8-stable is 8.8.7+ce.1-0-g51164768

Do not use container_registry and registry_* options

This does almost (*) only pure merge. We will slaposify / adjust config
and corresponding md5sum in the following patches.

gitlab_registry_http_config are put in comment

Conflicts:
	software/gitlab/template/gitlab.yml.in
	software/gitlab/template/nginx.conf.in
	software/gitlab/template/rack_attack.rb.in
	software/gitlab/template/smtp_settings.rb.erb

/reviewed-by: TrustMe

Add a class SlapConfigParser which overrite ConfigParser.write method to fix parse problem when
configuration like:
[section]
foo += bar

is included in buildout file. softwaretype recipe will generate
buildout file with foo + = bar because ConfigParser doesn't reconize +=
delimiter and read key as "foo +", value as "bar".
Then ConfigParser.write method generate

[section]
foo + = bar
...

This is invalid with buildout version 2

/reviewed-on nexedi/slapos!100

@Tyagov, please review.

Now the test runners use the LD_PRELOAD environment variable to load
the correct libstdc++ built by slapos and restart themselves.

The restart is necessary because LD_PRELOAD is only taken into account
at the time the process is started. Changing this environment variable
in runtime has no effects.

Tests results are available at Nexedi's ERP5 and are 100% ok: https://nexedi.erp5.net/test_result_module/20160927-3FA05624.

/reviewed-on nexedi/slapos!99

Document how to setup replication.
Add support for no-zope ERP5 instance, so it is possible to setup a
replicating, full-featured ERP5 Cluster instance tree without providing
service until it needs to take over its upstream, or to become
otherwise independent.

Buildout expects all needed sections to have been accessed during the
sectoin initialisation pass. It does not like having them accessed during
the "install" pass, and emits a very cryptic error:
  While:
    Installing publish.
  Error: Missing option: publish-early:__buildout_signature__

  Version up only goes to buildout2 case, buildout 1 series keep the same.

For sftp to work we need to define sftp "subsystem" on server side in
sshd speak. Sftp support in turn allows for sshfs to work.

NOTE

	Subsystem sftp internal-sftp

won't work because we are doing games with ForceCommand and reexecution
of "$SSH_ORIGINAL_COMMAND" (see previous patch for details).

2a733418 (slaprunner: replaces dropbear by openssh) switched slaprunner
sshd from dropbear to openssh (thanks for it once again) and used
ForceCommand to imitate login shell to be partitions `${bash:location}/bin/bash -l`.

For case when original ssh command is explicitly provided
("$SSH_ORIGINAL_COMMAND" is not empty) the code there tries to just
execute what was provided by caller.

However the reexecution is not completely correct as for cases when
origin command contains some quoting, 1 level of quotes is removed.

The case when I hit this in practice is git access to repositories over
ssh. When git wants to e.g. clone a repository it runs `git-upload-pack
quoted-path-to-repo.git` and fails this way:

```
$ GIT_TRACE=1 git clone ssh://host1/~/B.git
19:28:23.558858 git.c:350               trace: built-in: git 'clone' 'ssh://host1/~/B.git'
Cloning into 'B'...
19:28:23.572994 run-command.c:336       trace: run_command: 'ssh' 'host1' 'git-upload-pack '\''~/B.git'\'''
fatal: ''~/B.git'' does not appear to be a git repository
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
```

Quoting SSH_ORIGINAL_COMMAND and evaluating it completely via eval fixes this.

/cc @Nicolas, @alain.takoudjou

Because this url is use by [erp5_promise](https://lab.nexedi.com/nexedi/erp5/blob/450ae11128503934b4d81de8c7c72ba7aa7948d5/bt5/erp5_promise/SkinTemplateItem/portal_skins/erp5_promise/Alarm_checkPromiseConversionServer.py#L5) to configure cloudooo in the preference, but erp5 preference is not supposed to be `cloudooo://` but just `http://` . 

FYI, this erp5_promise is the **Check Site Consistency** part described in the "ERP5 on webrunner" installation procedure as documented in https://www.erp5.com/erp5-HowTo.Request.Erp5.Instance.On.Slap.Os.Webrunner

/cc @tc @vpelletier 

/reviewed-on nexedi/slapos!94

It depends on RINA libraries installed system-wide, which in turn depend on
other libraries like libssl and zlib.

Slapos.toolbox has a new dependency to pycurl, which fails to compile if libcurl cannot be found. This commit adds the component/pycurl dependency to most of the Software Releases which have a dependency to slapos.toolbox.

I tried to do it globally, so I couldn't check the compilation of all the modified SRs. Is it a way to test that all SRs still compiled ?

Also, did I forget some important SR ?

@rafael @alain.takoudjou @kazuhiko @jm 

/reviewed-on nexedi/slapos!97