It allows to cache on 4 levels:
a) tree: for all instances/domains runining on a certain source path
b) instance: for all sessions on a given instance
c) session: for all requests of a session, same as egw_session::appsession()
d) request: just for this request (same as using a static variable)
There's a get, a set and a unset method for each level: eg. getTree()
or setInstance(), as well as a variant allowing to specify the level as first
parameter: eg. unsetCache()
getXXX($app,$location,$callback=null,array $callback_params,$expiration=0)
has three optional parameters allowing to specify:
3. a callback if requested data is not yes stored. In that case the
callback is called and it's value is stored in the cache AND retured
4. parameters to pass to the callback as array, see call_user_func_array
5. an expiration time in seconds to specify how long data should be cached,
default 0 means infinit (this time is not garantied and not
supported for all levels!)
Data is stored under an application name and a location, like
egw_session::appsession().
In fact data stored at cache level egw_cache::SESSION, is stored in
the same way as egw_session::appsession() so both methods can be used
with each other.
The $app parameter should be either the app or the class name, which
both are unique.
The tree and instance wide cache uses a certain provider class, to
store the data eg. in memcached or if there's nothing else configured
in the filesystem (eGW's temp_dir).
using octal numbers with mysql leads to funny results:
select 384 & 0400 --> 384 not 256=0400
--> converted 0400, 040 and 04 to 256, 32 and 4 for mysql"
- egw_vfs::stat and egw_vfs_stream_wrapper::url_stat now both have a
parameter $try_create_home=false, which do not create a non-existing
home-directory by default.
- filemanger_ui calls egw_vfs::stat($path,true) to create an evtl.
missing home dir (in case it does not exist because of previous
problems)
--> fixes not working home-dir creation or rename, because url_stat
already tried to create the home-dir
and GroupDAV. The "sessionid" get's constructed from the basic auth
credentials and is not random (as the clients dont store them).
--> speeds up the use of *DAV
--> stops *DAV handlers to created numerious sessions
- fgetcsv only works correct, if setlocal is called with an existing and
correct local
- improved projectmanager method guess_local and moved it to
common::setlocal, which takes now the charset, lang and country of the
user into account
- csv-import also displays now the conversation done and reads usernames
in brackets
- added some missing fields
- all: false (default) = ignore files starting with a dot '.',
true = show all files (. and .. are always ignored!)
- exec: false (default) = do NOT allow to upload or modify scripts,
true = allow it (if docroot is mounted, this allows to run scripts!)
--> deny_script method was added to egw_vfs and calls to it from
filemanager
Other fixes:
- missing write rights of the webserver were not removed from perms
(causing warnings to be displayed in the ui)
- rename was not working due to typos
--> should be backported for obvious reasons to 1.6
- symlinks are now read by url_stat and dir_opendir and stored in the stat cache, to minimize DB accesses
- negative url_stat calles (not found) are now stored in the stat cache too, to minimize DB access
- added new log level 3, which adds method-name and line number to the sql statements"
- sqlfs can store now symlinks (implements symlink and readlink)
- vfs resolves symlinks before calling a mounted stream-wrapper
--> symlinks can be between different mount-points
- filemanger can create symlinks and follows them
- etemplate vfs_widget displays symlinks (to be improved)
Happy testing :-)
--> now you can find the error in the error_log and dont get only a blank page
(also optimized it so far, that we first try to autoload the class and use the diverse \"magic\" only if that fails)"
per session and then operate on that cache.
This deliminates ~25 database queries for each sitemgr page.
Test carefully before applying this to a production sytem!
- change the processing of slowsync, to use the content_map instead of
trying to build a new one. This caused duplication issues on the
client if multiple similar records where stored, because only the first
one found in the server-db was matched, These duplicate entries at client
side had no entry at serverside, so deleting the wrong one
on the client (the content with a valid map entry) could cause
unwanted data loss at server side, because it is impossible for the
user to see what is a duplicate, and what is not.
see also:
http://www.nabble.com/again---syncml-duplication-issue-to20333619s3741.html
- reenabled UID from syncml clients, because it was partly used this caused
issues during SlowSync if the content was changed.
- infolog, calendar if a uid is found in the provided data, allway try to
find the corresponding content first using only the UID, instead of
using the content-id taken from content_map.
also fixed:
- a few fixes in ./notes
- creating an entry on the client that can not be imported,
(Example, Nokia E Series Appointment without a Title)
will no longer create an invalid content-map entry
However, at client side this is still counted in the Protocol as
Server-Add
manufacturer and the recogniced GroupDAV client as product name.
This way we are able to handle different GroupDAV clients, as we
allready do with different SyncML clients.
Also removed the no longer needed code enabling the use of the real UID,
as SyncML does no longer misuse the UID for it's GUID.
longer use GUIDs containing eGW's install_id, as the information is
irrellevant for SyncML and cause doublications of entries if the
install_id changes.
I plan to have a new rc4 Wednesday or Thursday containing these changes.
- adding the application ('syncml')
- replacing next_record()/f() with fetch()/fetchSingle() or looping over the result object
Thanks to Philip Herbert from Knauber for testing it"
- exceptions get now always logged to the error_log
- in the webgui it's now configurable, if the message contains a
stacktrace (incl. function arguments) - default no (security)
- command line interfaces get detected and contain no html anymore
- webdav and groupdav send the exceptions as basic auth realms to the
client
- webdav and groupdav login failures contain the reason as part of the
basic auth realm
- egw_vfs::download_url as not encoding + or ' ' in pathes
- HTTP_WebDAV_Server was urldecoding $_SERVER[PATH_INFO], which is
wrong, as it is NOT encoded
- HTTP_WebDAV_Server was NOT urlencoding the pathes in PROPFIND
responses, causing eg. cadaver not to be able to use dirs containing
+ or space
sqlfs stores files with fs_id < 100 directly under /sqlfs in the files
dir. They conflict with directories created for fs_id >= 1000.
--> fs_id < 100 are now in a directory /sqlfs/00
You need to run the 1.5.016 update or you will not find the content of
files with fs_id < 100 anymore!
reads of entries from the database: Applications can call
egw_link::set_cache($app,$id,$title,$file_access=null)
from their search or read method, to eliminate the need to query the
entries again, when the egw_link class, link widget or links stream wrapper
needs title or file_access values later.
This offloads the caching to the link class, and improves performance a
lot, specially for infolog.
The cache is stored in the session and modified or deleted items get
removed, when the link class get notified about that anyway.
- removed include_root (is now always identical to server_root)
- removed mcrypt_version (not relevant to todays php versions)
- replaced session_type with session_handler
- look is now a bit more like setup and eGroupware itself
--> updated header-version to 1.29 forcing everyone to update
- flag in session if it is encrypted to prevent calling the encryption more then once, which stalls the session-content
- egw_session::session_comit() method calls now encrypt() too, as it closes the session, before the destructor is called
- hack to fix PHP Fatal error: Cannot use string offset as an array, which happens sometime in felamimail under php5.2
- some more docu"
-documenting the needed skripting for the configured aspell path in egw_integration.txt
-re-imported spellcheck (aspell) integration from Jose
-switching aspell integration on in fckconfig.js, as documented in FCKeditor_2.x/Developers_Guide/Configuration/Spell_Checker
- it now also encrypts the egw object and egw_info array, stored in the session
- it no longer encrypts every egw_session::appsession() call, but the
whole array at once when the egw_session object gets destroyed
- mcrypt algo and mode are currently hardcoded to tripledes and ecb, as
we dont have the database connection, when they are needed. You can
add it as egw_info[server][mcrypt_{algo|mode}] in the header.inc.php
- fixed a bug, which let the session grow around 400k(!) each request
- if mcrypt or the selected algo/mode is not availible the session
encryption is switched off automatic, but an error is logged
this should fix the <br type="_moz"> bug.
-added the about button in all egw_config settings, as it enables the user to check the version of the editor
-added the show-blocks button in all egw_config settings, as it enables the user to check the structure elements of his document
+ path with clickable components
+ human readable size, mode, ...
+ mime icon with integrated thumbnail creation
- link widget uses now vfs-mime for it's icons
- thumbnail creation is now switched on with size 32px by default, it can
be switched of by the admin or user, in doing so explicitly
- mime-icons are moved from filemanager to etemplate, as not everyone
installs filemanager
- filemanager has now 3 display modi:
+ Current directory (with subdirs always on top)
+ Subdirs sorted in
+ Files from subdirs (shows recursive all files and you
can click on the path components thanks to new vfs widget)
a hashed directory structure based on the fs_id and not longer on the
path (which can not be recovered, once the filesystem get's corrupt)
--> Make backups (db AND files directory), before attempting the update !!!!!!!!
- the used storage (default filesystem) can be switched via a get-parameter in the url mounted (eg. sqlfs://default/?storage=db)
- please note the current (php5.2.6) problems:
a) retriving files via streams does NOT work for PDO_mysql (bindColum(,,PDO::PARAM_LOB) does NOT work, string returned)
(there's a workaround implemented, but it requires to allocate memory for the whole file!)
b) uploading/writing files > 1M fail on PDOStatement::execute() (setting PDO::MYSQL_ATTR_MAX_BUFFER_SIZE does NOT help)
(not sure if that's a bug in PDO/PDO_mysql or an accepted limitation)
--> now we need to implement an easy switch in setup to allow admins to use the db backend (does NOT require an directory outside the docroot)
currently you need to use filemanager/cli.php mount"
allowing to set an arbitrary uid, gid and mode for the vfs (like
mounting a dos Filesystem under Linux).
Can be used to mount eg. an upload dir writable only by Admins:
filemanger/cli.php mount --user root_admin --password whatever \
filesystem://dummy/var/www/html/uploads?group=Admins&mode=075 \
/uploads
Please note that you can NOT use filesystem:/something!
Fixed the vfs-classes to deal correctly with get-parameters used as
mount-options.
- new feature allowing apps to specify with index-page and icon to use
--> allows to install sitemgr-link icon without an extra application
(which was a symlink and got lost in every version upgrade)
PHP Fatal error: egw_db::connect(): The script tried to execute a method or access a property of an incomplete object ...
Problem was that session restore is now used regardless if it's explicitly switched on, egw_db::connect() only registered the required class, if it was explicitly switched on. Therefore it worked everywhere, where session restore was configured before, but failed everyone not used it before ;-)"
- DONT UPDATE ON A PROCUDTION SYSTEM (for the next few days)!
- eGW support from now on only php session handling
- custom session handlers (like the memcache one) can now be
implemented as classes and dont need to change any other code
- the class get's autoloaded and the name need to be configured
eg. in the header.inc.php as $egw_info[server][session_handler]
- session restore is now enabled by default (it's way faster and
works well with php5.1+)
- a db-bases session handler follows soon
- new GroupDAV v2 component-set attribute for collections
- getlastmodified & getcontentlength properties for infolog propfind
- fixed propfind on a single infolog entry to return just that entry
- getcontenttype of vevent and vtodo collection returns extra component
The problem seems to be line [784] of trunk/phpgwapi/inc/horde/Horde/iCalendar.php
$value = str_replace($this->_newline, '\n', $value);
When removing this line, the description value is correct on the client.
I could not find any sideffects during my tests, if some clients have
problems with this, I assume this would then have to be fixed at a higher
level, because the current state with this line just causes broken output.
From wikipedia regarding Linebreaks in QuotedPrintable:
If the data being encoded contains meaningful line breaks, they must be encoded as an ASCII CR LF sequence, not as their original byte values
Vcard extract without the reported line code:
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:Desc 1=0D=0ADesk 2=0D=0A=0D=0A
Vcard extract with the reported line of code
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:Desc 1\nDesk 2\n\n
pointed out by Philip Herbert. Carl Knauber Holding GmbH & Co KG
formal error.
This breaks sync for single contacts from egw to client.
example: photo in addressbook without blank line after the property value.
This way the devices are not compliant with RFC2426 (Vcard Version 3)
5. Differences From vCard v2.1
[...]
. Inline binary content must be "B" encoded and folded. A blank
line after the encoded binary content is no longer required.
[...]
This was pointed out by Philip Herbert. Carl Knauber Holding GmbH & Co KG
noticed while working on it:
- memory size error, when renaming a file after posting the list (eg.
clicking on home icon)
- renaming (moving) one file on an existing filename, put the file in an
inaccessible state
- renaming more then once, did not work
--> ToDo: add some ajax to notify the user, when he tries to overwrite
an other file while renaming one
