Config class refactoring

[vipsoft]

Increase abstraction and reduce runtime dependence on Zend Framework. This provides a more uniform interface for core and plugin developers.

That is:

• use Piwik_Config::getInstance() instead of Zend_Registry::get('config')
• use Piwik_Config::getInstance() instead of Piwik_Tracker_Config::getInstance()

Read individual setting or entire section:

$value = Piwik_Config::getInstance()->section['option'];

$section = Piwik_Config::getInstance()->section;

Change individual setting or entire section:

Piwik_Config_Writer::getInstance()->section = $newSection;

Piwik_Config_Writer::getInstance()->section['option'] = $value;

Split/move core/Config.ini.php into:

• Config/Abstract.php - base class
• Config.php - config reader; can set but does not save
• Config/Writer.php - config reader/writer; will save changes

Remove core/Config/Ini.php and core/Tracker/Config.php.

Impact:

• not backwards compatible because Zend_Config isn't used to represent array options
• 3rd party plugins affected: TBD

Must prevent recurrence of #1945. (Ideally, repair config.ini.php, perhaps by comparing against global.ini.auto-backup-before-update.php to isolate which General settings were actually custom.)

[vipsoft]

Another thought: could we drop the .ini format and switch to pure php? We could use var_export when writing. This allows us to eliminate the dependency on parse_ini_file() -- and the workaround when it is disabled. Loading would be faster and it is cacheable by APC.

[matt]

Refactoring sounds great. However I vote for keeping INI format for now, as it is universal. Editing php arrays for config is not as friendly, and also wouldn't easily support comments.

[vipsoft]

(In [5484]) refs #1713

[vipsoft]

Quick grep shows the following non-core plugins are affected:

• CASLogin
• ClickHeat
• DoNotTrack
• EntryPage
• Gall
• GeoIPMap
• HttpAuthLogin
• KSVisitorImport
• LdapLogin
• Maps
• MobileAnalytics
• OpenIDLogin
• OxidPlugin
• Signup

[vipsoft]

(In [5537]) refs #1713 - create branch for review

[matt]

How do you plan to deal with the affected plugins? will you repackage all plugins to solve the bug and require an update?

I wouldn't put high priority since it won't add any feature, fix a bug and it breaks all plugins.

Otherwise looks good, always nice to simplify code base!

[vipsoft]

This would probably be a good time to fix/repackage all the affected plugins since 3rd party plugins aren't routinely tested for compatibility.

Also:

• blog and/or post to piwik-hackers before this makes it into a release
• an update script would disable known affected 3rd party plugins

[vipsoft]

(In [5586]) refs #1713 - to matt for review/feedback

[matt]

Wow, nice work Anthon! long due change indeed :) Review:

• isMacOs && Ip.test.php changes are probably unrelated to this change
• Did you check that password in config files still work when it contains <>&"' characters (I saw that html decode call was removed in plugins/DBStats/API.php

• To go forward with this proposal, we must first ensure that all plugins are updated and packaged for the new version.
  • OR, what do you think about making the old format Zend_Registry::get('config')->General->anonymous_user_enable_use_segments_API still working for "READ" operations? Ie. we could add a fake config object that would reroute to the Piwik_Config::getInstance() equivalent? I suppose that by doing so, most plugins would not need repackaging since most plugins would only READ values from the file. The goal would be to make this commit less painful and less risky and not require all existing plugin users to upgrade their third party plugins after the piwik core update. What do you think?

[vipsoft]

• Yes, there are some unrelated changes there. It just happens that my dev environment was MacOS X. I'll merge the unrelated changes into trunk this weekend.
• DBStats/API.php was double decoding. It predates [1165].

I was trying to avoid cruft in the code base, but I agree a backward compatibility layer would reduce headaches. I'll look into this.

[vipsoft]

in [5608], added backward compatibility layer

[matt]

Great :) After adding this layer, what are the plugins that still need updating? is there any at all? If so, can we scheduled this for Piwik 1.8? Can you think of any other risk in this change? Thanks!!

[vipsoft]

Should work with the aforementioned plugins. I'll be busy testing this weekend. Otherwise looking good for 1.8

[vipsoft]

(In [5615]) refs #1713 - naming inconsistency with Config.php

[matt]

See also the double quotes bug in #2968

[vipsoft]

(In [5951]) refs #1713 - merge dev branch to trunk (config class refactoring)

[vipsoft]

(In [5952]) refs #1713 - fix svn property on core/Config/Compat.php

[vipsoft]

(In [5954]) refs #1713 - partially resolve regression from the merge

[vipsoft]

(In [5955]) refs #1713, refs #1331 - workaround for auto-update

[vipsoft]

(In [5957]) refs #1713 - rollback this file to r5454; workaround for #1331

[matt]

(In [5959]) Refs #1713 Fixing NOTICE when element not in config file

[matt]

Great code cleanup and getting rid of akward code for sure :)

I notice that the trunk is broken, many integration tests fail - do you see the same behavior on your local box?

for some reason Jenkins fails at webtests when it should fail in the unit/integration tests, not sure why?

Thanks

[matt]

(In [5960]) Refs #1713 fixes changing SU email

[matt]

(In [5961]) Refs #1713 Admin settings were not saved

[vipsoft]

I don't yet know why the integration tests pass on dev6 but fail locally.

[matt]

OK, I can help investigate in ~ 1 hour if you're still stuck!

[matt]

• On rewriting the config file, the sections are not put back in the order they were in. It is important that sections are written in the order found in global.ini.php (ie. superuser, then database, etc.)

[matt]

• Running integration tests is overwriting the local config.ini.php !! not so good, I lost my local mods

EDIT: Strange I just lost my [Tracker] section. Thanks to Eclipse local history I could restore :)

[matt]

• Install is broken too..: No entry is registered for key 'config'

[vipsoft]

(In [5967]) refs #1713 - fix installation; will have to workaround #1331 some other way

[vipsoft]

(In [5968]) refs #1713 - splitting reader/writer just isn't possible with the way integration tests run

[vipsoft]

(In [5971]) refs #1713 - fix installation and workaround #1331 (deja vu?)

[vipsoft]

To do:

• broken one click update; I'll narrow this down today
• re-write sections in original order, appending new sections at the end
• investigate disappearing section (was it the same as global.in?)

[vipsoft]

(In [5980]) refs #1713, refs #1331 - this should fix the one click update

[vipsoft]

(In [5987]) refs #1713

[matt]

(In [6000]) Refs #1713

• keep sections that don't exist inglobal.ini.php (prevent NOTICE).
• save sections in order that they appear in global.ini.php

[matt]

(In [6002]) Refs #1713

• Ignoring local config.ini.php when running tests
• Disable IP Anonymisation if it's enabled on dev box (was causing some integration tests fail)

[vipsoft]

(In [6011]) refs #1713 - refactor for testability

[vipsoft]

(In [6012]) refs #1713 - add unit test for dumpConfig()

[vipsoft]

(In [6021]) refs #1713 - update the dirty bit and sorting tests

[vipsoft]

Four of the tests are failing.

• local copy (different), cached get - this is suboptimal
• local copy, cached set local, new section - this is a bug
• local copy, cached set global, new section - this is a bug
• sort, common sections before new section - this is a bug

[vipsoft]

(In [6022]) refs #1713 - aha, bad tests

[vipsoft]

(In [6023]) refs #1713 - fix problem with new sections not being added; comment out two tests where writer is sub-optimal (i.e., it writes out a new config file even if there is no change)

[matt]

The web tests do not fail, but trunk install is broken currently

"SQLSTATE[42S02]: Base table or view not found: 1146 Table 'piwik_trunk_testsvn1.piwik_option' doesn't exist"

[vipsoft]

(In [6028]) refs #1713 - remove Config destruct()

[vipsoft]

(In [6029]) refs #1713 - set() should set dirty bit

[vipsoft]

(In [6034]) refs #1713 - not so lightweight anymore, but still an improvement by unifying Zend_Config and Piwik_Tracker_Config

[matt]

Feedback

• Great work. Maintaining BC and simplifying code and interfaces, big win for the software maintainability and plugin developers.
• Do you know why the webtest was not failing when trunk was broken? Is it because the webtest was disabled maybe?
• It looks good for me, is the ticket now fixed or are there pending tasks?

Thanks for your nice follow up too, that was a tricky large core change :)

[matt]

I noticed a bug:

• when the config.ini.php contains a value the same as global.ini.php, it gets deleted at the next config file write.
• this is causing some frustration, because it means opening the file global.ini.php every time again to enable some custom features
• the solution would be that once a value has been modified in config.ini.php, it is always left there, even when the value is set to the same as in global.ini.php

[vipsoft]

It's intentional and is a result of array_unmerge.

[matt]

I think it's more usable if we leave the values in the file... it's common to switch 1 to 0 to 1 etc. if the line has to be fully written each time, it's not practical. I vote for revert and do it like before, thoughts?

[vipsoft]

This would reintroduce #1945 where changing a local setting programmatically would copy the rest of the section.

[matt]

OK, maybe we can keep track of which config elements were first read from config.ini.php, keep a flag for those, and always write them out in config.ini.php?

[matt]

(In [6075]) refs #1713 removing login=root from global.ini.php or it is removed from config.ini.php on rewrite if SU login is root

[vipsoft]

(In [6109]) refs #1713 - workaround php 5.1.x - php 5.2.0 reference bugs when config is returned by reference

[vipsoft]

(In [6111]) refs #1713 - tracker config did not previously throw exceptions (BC); revert [6107] and part of [6103]

[matt]

Thanks for the better fix!!

[matt]

Open critical issues related to the new config:

• login=root and salt=XX are not in the config file on rewrite. During the update to 1.7.2-rc the login=root and salt=xx should be copied to the config./ini.php (or it prevents Super User login!! CRITICAL)
• for usability it is important to leave the modified config settings in config.ini.php so as to avoid users to go back and forth between config and global files

@vipsoft Do you think these could be tackled in next days as to be able to release 1.7.2 ? Thanks!!

[matt]

• Maybe it would be possible to remove Zend/Config files from libs/Zend ?

[vipsoft]

Replying to matt:

Open critical issues related to the new config: * login=root and salt=XX are not in the config file on rewrite. During the update to 1.7.2-rc the login=root and salt=xx should be copied to the config./ini.php (or it prevents Super User login!! CRITICAL) * for usability it is important to leave the modified config settings in config.ini.php so as to avoid users to go back and forth between config and global files @vipsoft Do you think these could be tackled in next days as to be able to release 1.7.2 ? Thanks!!

I think I understand the first issue. A core/Updates/ script should suffice.

I understand the potential usability issue in the second case, but I can't think of any easy fix at the moment.

Replying to matt:

* Maybe it would be possible to remove Zend/Config files from libs/Zend ?

Off-hand, I'm going to say yes it's possible, but I haven't tried.

[matt]

A core/Updates/ script should suffice.

OK

I understand the potential usability issue in the second case, but I can't think of any easy fix at the moment.

My idea was to remove the array_unmerge, and instead, write out only the settings that are currently written out in the local config.ini.php (not remove any setting or add any new one from global).

eg. This could be done by checking configLocal before when writing out the config file?

[vipsoft]

array_unmerge solves the problem of #1945, where changing one setting in a section would copy the rest of the section, such that the entire section was duplicated in local config.ini.php.