/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ /* * This file is part of the LibreOffice project. * * This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. * * This file incorporates work covered by the following license notice: * * Licensed to the Apache Software Foundation (ASF) under one or more * contributor license agreements. See the NOTICE file distributed * with this work for additional information regarding copyright * ownership. The ASF licenses this file to you under the Apache * License, Version 2.0 (the "License"); you may not use this file * except in compliance with the License. You may obtain a copy of * the License at http://www.apache.org/licenses/LICENSE-2.0 .
*/
/** After the fact documentation - hopefully it is correct. * * NOTE TO DEVELOPERS: instsetoo_native/ooenv sets the environment variable OOO_DISABLE_RECOVERY * which means that your non-gdb execution of the compiled code will not display the recovery dialog * and if officecfg::Office::Recovery::RecoveryList has some autosave files to recover, * then autorecovery itself will not be started, and none of the code in this file will be executed. * THEREFORE you probably want to COMMENT OUT that environment variable when working in this file. * * AutoRecovery handles 3 types of recovery, as well as periodic document saving * 1a) timed, automatic saving of the documents (aka UserAutoSave) * or more commonly: * 1b) timed, ODF, temporary recovery files created in the backup folder (default setting) * -temporary: deleted when the document itself is saved * -handles the situation where LO immediately exits (power outage, program crash, pkill -9 soffice) * -not restored immediately (user needs to restart LibreOffice) * -no guarantee of availability of recovery file (since deleted on document save) * or original document (perhaps /tmp, removeable disk, disconnected server). * -TODO tdf#57414: if SessionSave not desired, don't recover unmodified files. * 2) emergency save-and-restart immediately triggers creation of temporary, ODF, recovery files * -handles the situation where LO is partially functioning (pkill -6 soffice) * -restore attempted immediately, so try to restore entire session - all open files * -always create recovery file for every open document in emergency situation * -works without requiring AutoRecovery to be enabled * 3) session save on exit requested by OS or user * -triggered by OS's shutdown/logout * -appears to be purely theoretical: no known working OS at the moment. * -also no known way for user to initiate within LO (tdf#146769) * -same as emergency save, except maybe more time critical - OS kill timeout * -not restored until much later - the user has stopped doing computer work * -always create recovery file for every open document: needed for /tmp, disconnected docs * * All of these use the same recovery dialog - re-opening all the files listed in the RecoveryList * of the user's officecfg settings. * * Since these 3 have very different expectations, and yet share the same code, keep all of them * in mind when making code changes. * * Note: often, entries in m_lDocCache are copied. So realize that changes to aInfo/rInfo might not * apply to async events like mark-document-as-saved-and-delete-TMP-URLs or set-modified-status, * or ignoreClosing, or ListenForModify. For example, DocState::Modified should be considered only * a good hint, and not as definitively accurate.
*/
namespace {
/** @short hold all needed information for an asynchronous dispatch alive.
@descr Because some operations are forced to be executed asynchronously (e.g. requested by our CrashSave/Recovery dialog) ... we must make sure that this information won't be set as "normal" members of our AutoRecovery instance. Otherwise they can disturb our normal AutoSave-timer handling. e.g. it can be unclear then, which progress has to be used for storing documents...
*/ struct DispatchParams
{ public:
DispatchParams();
DispatchParams(const ::comphelper::SequenceAsHashMap& lArgs , const css::uno::Reference< css::uno::XInterface >& xOwner);
void forget();
public:
/** @short can be set from outside and is provided to our internal started operations.
@descr Normally we use the normal status indicator of the document windows to show a progress. But in case we are used by any special UI, it can provide its own status indicator object to us - so we use it instead of the normal one.
*/
css::uno::Reference< css::task::XStatusIndicator > m_xProgress;
/** TODO document me */
OUString m_sSavePath;
/** @short define the current cache entry, which should be used for current
backup or cleanUp operation ... which is may be done asynchronous */
sal_Int32 m_nWorkingEntryID;
/** @short used for async operations, to prevent us from dying.
@descr If our dispatch() method was forced to start the internal operation asynchronous... we send an event to start and return immediately. But we must be sure that our instance live if the event callback reach us. So we hold a uno reference to ourself.
*/
css::uno::Reference< css::uno::XInterface > m_xHoldRefForAsyncOpAlive;
};
/** These values are used as flags and represent the current state of a document. Every state of the life time of a document has to be recognized here.
@attention Do not change (means reorganize) already used numbers. There exists some code inside SVX, which uses the same numbers, to analyze such document states. Not the best design ... but may be it will be changed later .-)
*/ enumclass DocState: sal_Int32
{ /* TEMP STATES */
/// default state, if a document was new created or loaded
Unknown = 0, /// modified against the original file
Modified = 1, /// an active document can be postponed to be saved later.
Postponed = 2, /// was already handled during one AutoSave/Recovery session.
Handled = 4, /** an action was started (saving/loading) ... Can be interesting later if the process may be was interrupted by an exception. */
TrySave = 8,
TryLoadBackup = 16,
TryLoadOriginal = 32,
/* FINAL STATES */
/// the Auto/Emergency saved document is not usable any longer
Damaged = 64, /// the Auto/Emergency saved document is not really up-to-date (some changes can be missing)
Incomplete = 128, /// the Auto/Emergency saved document was processed successfully
Succeeded = 512
};
/** implements the functionality of AutoSave and AutoRecovery of documents - including features of an EmergencySave in case a GPF occurs.
*/ typedef ::cppu::WeakComponentImplHelper<
css::lang::XServiceInfo,
css::frame::XDispatch,
css::document::XDocumentEventListener, // => css.lang.XEventListener
css::util::XChangesListener, // => css.lang.XEventListener
css::util::XModifyListener > // => css.lang.XEventListener
AutoRecovery_BASE;
class AutoRecovery : private cppu::BaseMutex
, public AutoRecovery_BASE
, public ::cppu::OPropertySetHelper // => XPropertySet, XFastPropertySet, XMultiPropertySet
{ public: /** @short indicates the results of a FAILURE_SAFE operation
@descr We must know, which reason was the real one in case we couldn't copy a "failure document" to a user specified path. We must know, if we can forget our cache entry or not.
*/ enum EFailureSafeResult
{
E_COPIED,
E_ORIGINAL_FILE_MISSING,
E_WRONG_TARGET_PATH
};
// TODO document me enum ETimerType
{ /** the timer should not be used next time */
E_DONT_START_TIMER, /** timer (was/must be) started with normal AutoSaveTimeIntervall */
E_NORMAL_AUTOSAVE_INTERVALL, /** timer must be started with special short time interval,
to poll for an user idle period */
E_POLL_FOR_USER_IDLE, /** timer must be started with a very(!) short time interval,
to poll for the end of an user action, which does not allow saving documents in general */
E_POLL_TILL_AUTOSAVE_IS_ALLOWED, /** don't start the timer - but calls the same action then before immediately again! */
E_CALL_ME_BACK
};
/** @short combine different information about one office document. */ struct TDocumentInfo
{ public:
/** @short points to the document. */
css::uno::Reference< css::frame::XModel > Document;
/** @short knows, if the document is really modified since the last autosave, or was postponed, because it was an active one etcpp...
@descr Because we have no CHANGE TRACKING mechanism, based on office document, we implements it by ourself. We listen for MODIFIED events of each document and update this state flag here.
Further we postpone saving of active documents, e.g. if the user works currently on it. We wait for an idle period then...
*/
DocState DocumentState;
/** Because our applications not ready for concurrent save requests at the same time, we have suppress our own AutoSave for the moment, a document will be already saved by others.
*/ bool UsedForSaving;
/** For every user action, which modifies a document (e.g. key input) we get a notification as XModifyListener. That seems to be a "performance issue" .-) So we decided to listen for such modify events only for the time in which the document was stored as temp. file and was not modified again by the user.
*/ bool ListenForModify;
/** For SessionSave we must close all open documents by ourself. But because we are listen for documents events, we get some ... and deregister these documents from our configuration. That's why we mark these documents as "Closed by ourself" so we can ignore these "OnUnload" or disposing() events .-)
*/ bool IgnoreClosing;
OUString OldTempURL; // previous recovery file (filename_0.odf) which will be removed
OUString NewTempURL; // new recovery file (filename_1.odf) that is being created
OUString AppModule; // e.g. com.sun.star.text.TextDocument - used to identify app module
OUString FactoryService; // the service to create a document of the module
OUString RealFilter; // real filter, which was used at loading time
OUString DefaultFilter; // supports saving of the default format without losing data
OUString Extension; // file extension of the default filter
OUString Title; // can be used as "DisplayName" on every recovery UI!
css::uno::Sequence< OUString >
ViewNames; // names of the view which were active at emergency-save time
sal_Int32 ID;
};
/** @short used to know every currently open document. */ typedef ::std::vector< TDocumentInfo > TDocumentList;
// member
private:
/** @short the global uno service manager. @descr Must be used to create own needed services.
*/
css::uno::Reference< css::uno::XComponentContext > m_xContext;
/** @short points to the underlying recovery configuration. @descr This instance does not cache - it calls directly the configuration API!
*/
css::uno::Reference< css::container::XNameAccess > m_xRecoveryCFG;
/** @short proxy weak binding to forward Events to ourself without an ownership cycle
*/
css::uno::Reference< css::util::XChangesListener > m_xRecoveryCFGListener;
/** @short points to the used configuration package or.openoffice.Setup @descr This instance does not cache - it calls directly the configuration API!
*/
css::uno::Reference< css::container::XNameAccess > m_xModuleCFG;
/** @short holds the global event broadcaster alive, where we listen for new created documents.
*/
css::uno::Reference< css::frame::XGlobalEventBroadcaster > m_xNewDocBroadcaster;
/** @short proxy weak binding to forward Events to ourself without an ownership cycle
*/
css::uno::Reference< css::document::XDocumentEventListener > m_xNewDocBroadcasterListener;
/** @short because we stop/restart listening sometimes, it's a good idea to know if we already registered as listener .-)
*/ bool m_bListenForDocEvents; bool m_bListenForConfigChanges;
/** @short for an asynchronous operation we must know, if there is at least one running job (may be asynchronous!).
*/
Job m_eJob;
/** @short the timer, which is used to be informed about the next saving time ... @remark must lock SolarMutex to use
*/
Timer m_aTimer;
/** @short make our dispatch asynchronous ... if required to do so! */
std::unique_ptr<vcl::EventPoster> m_xAsyncDispatcher;
/** @short indicates, which time period is currently used by the internal timer.
*/
ETimerType m_eTimerType;
/** @short this cache is used to hold all information about recovery/emergency save documents alive.
*/
TDocumentList m_lDocCache;
// TODO document me
sal_Int32 m_nIdPool;
/** @short contains all status listener registered at this instance.
*/
comphelper::OMultiTypeInterfaceContainerHelperVar3<css::frame::XStatusListener, OUString> m_lListener;
/** @descr This member is used to prevent us against re-entrance problems. A mutex can't help to prevent us from concurrent using of members inside the same thread. But e.g. our internally used stl structures are not threadsafe ... and furthermore they can't be used at the same time for iteration and add/remove requests! So we have to detect such states and ... show a warning. May be there will be a better solution next time ... (copying the cache temp. bevor using).
And further it's not possible to use a simple boolean value here. Because if more than one operation iterates over the same stl container ... (only to modify it's elements but don't add new or removing existing ones!) it should be possible doing so. But we must guarantee that the last operation reset this lock ... not the first one ! So we use a "ref count" mechanism for that."
*/
sal_Int32 m_nDocCacheLock;
/** @descr These members are used to check the minimum disc space, which must exists to start the corresponding operation.
*/
sal_Int32 m_nMinSpaceDocSave;
sal_Int32 m_nMinSpaceConfigSave;
// css.document.XDocumentEventListener /** @short informs about created/opened documents.
@descr Every new opened/created document will be saved internally so it can be checked if it's modified. This modified state is used later to decide, if it must be saved or not.
@param aEvent points to the new created/opened document.
*/ virtualvoid SAL_CALL documentEventOccured(const css::document::DocumentEvent& aEvent) override;
private: virtualvoid SAL_CALL disposing() final override;
/** @short open the underlying configuration.
@descr This method must be called every time a configuration call is needed. Because method works together with the member m_xCFG, open it on demand and cache it afterwards.
@throw [com.sun.star.uno.RuntimeException] if config could not be opened successfully!
@threadsafe
*/ void implts_openConfig();
/** @short read the underlying configuration.
@descr After that we know the initial state - means: - if AutoSave was enabled by the user - which time interval has to be used - which recovery entries may already exists
@throw [com.sun.star.uno.RuntimeException] if config could not be opened or read successfully!
@threadsafe
*/ void implts_readConfig();
/** @short read the underlying configuration...
@descr ... but only keys related to the AutoSave mechanism. Means: State and Timer interval. E.g. the recovery list is not addressed here.
@throw [com.sun.star.uno.RuntimeException] if config could not be opened or read successfully!
@threadsafe
*/ void implts_readAutoSaveConfig();
/** After the fact documentation * @short adds/updates/removes entries in the RecoveryList - files to be recovered at startup * * @descr Deciding whether to add or remove an entry is very dependent on the context! * EmergencySave and SessionSave are interested in all open documents (which may not * even be available at next start - i.e. /tmp files might be lost after a reboot, * or removable media / server access might not be connected). * TODO: On the other hand, timer-based autorecovery in theory * would not be interested in recovering the session, * but only modified documents that are recoverable. * (TODO: unless the user always wants to recover a session).
*/ void implts_flushConfigItem(AutoRecovery::TDocumentInfo& rInfo, bool bRemoveIt = false, bool bAllowAdd = true);
// TODO document me void implts_startListening(); void implts_startModifyListeningOnDoc(AutoRecovery::TDocumentInfo& rInfo);
// TODO document me void implts_stopListening(); void implts_stopModifyListeningOnDoc(AutoRecovery::TDocumentInfo& rInfo);
/** @short stops and may be(!) restarts the timer.
@descr A running timer is stopped every time here. But starting depends from the different internal timer variables (e.g. AutoSaveEnabled, AutoSaveTimeIntervall, TimerType etcpp.)
@throw [com.sun.star.uno.RuntimeException] if timer could not be stopped or started!
@threadsafe
*/ void implts_updateTimer();
/** @short stop the timer.
@descr Double calls will be ignored - means we do nothing here, if the timer is already disabled.
@throw [com.sun.star.uno.RuntimeException] if timer could not be stopped!
/** @short validate new detected document and add it into the internal document list.
@descr This method should be called only if it's clear that a new document was opened/created during office runtime. This method checks if it's a top level document (means not an embedded one). Only such top level documents can be recognized by this auto save mechanism.
@param xDocument the new document, which should be checked and registered.
/** @short remove the specified document from our internal document list.
@param xDocument the closing document, which should be deregistered.
@param bStopListening sal_False: must be used in case this method is called within disposing() of the document, where it makes no sense to deregister our listener. The container dies... sal_True : must be used in case this method is used on "deregistration" of this document, where we must deregister our listener .-)
// TODO document me void implts_markDocumentModifiedAgainstLastBackup(const css::uno::Reference< css::frame::XModel >& xDocument);
// TODO document me void implts_updateModifiedState(const css::uno::Reference< css::frame::XModel >& xDocument);
// TODO document me void implts_updateDocumentUsedForSavingState(const css::uno::Reference< css::frame::XModel >& xDocument , bool bSaveInProgress);
// TODO document me void implts_markDocumentAsSaved(const css::uno::Reference< css::frame::XModel >& xDocument);
/** @short search a document inside given list.
@param rList reference to a vector, which can contain such document.
@param xDocument the document, which should be located inside the given list.
@return [TDocumentList::iterator] which points to the located document. If document does not exists - it's set to rList.end()!
*/ static TDocumentList::iterator impl_searchDocument( AutoRecovery::TDocumentList& rList , const css::uno::Reference< css::frame::XModel >& xDocument);
/** TODO document me */ void implts_changeAllDocVisibility(bool bVisible); void implts_prepareSessionShutdown();
/** @short save all current opened documents to a specific backup directory.
@descr Only really changed documents will be saved here.
Further this method returns a suggestion, if and how it should be called again. May be some documents was not saved yet and must wait for an user idle period ...
@param bAllowUserIdleLoop Because this method is used for different uses cases, it must know, which actions are allowed or not. Job::AutoSave => If a document is the most active one, saving it will be postponed if there exists other unsaved documents. This feature was implemented, because we don't wish to disturb the user on it's work. ... bAllowUserIdleLoop should be set to sal_True Job::EmergencySave / Job::SessionSave => Here we must finish our work ASAP! It's not allowed to postpone any document. ... bAllowUserIdleLoop must(!) be set to sal_False
@param pParams sometimes this method is required inside an external dispatch request. The it contains some special environment variables, which overwrites our normal environment. AutoSave => pParams == 0 SessionSave/CrashSave => pParams != 0
@return A suggestion, how the timer (if it's not already disabled!) should be restarted to fulfill the requirements.
/** @short save one of the current documents to a specific backup directory.
@descr It: - defines a new(!) unique temp file name - save the new temp file - remove the old temp file - patch the given info struct - and return errors.
It does not: - patch the configuration.
Note further: it patches the info struct more than ones. E.g. the new temp URL is set before the file is saved. And the old URL is removed only if removing of the old file was successful. If this method returns without an exception - everything was OK. Otherwise the info struct can be analyzed to get more information, e.g. when the problem occurs.
@param sBackupPath the base path for saving such temp files.
@param rInfo points to an information structure, where e.g. the document, its modified state, the count of autosave-retries etcpp. exists. It's used also to return the new temp file name and some other state values!
// TODO document me void implts_openOneDoc(const OUString& sURL ,
utl::MediaDescriptor& lDescriptor,
AutoRecovery::TDocumentInfo& rInfo );
// TODO document me void implts_generateNewTempURL(const OUString& sBackupPath ,
utl::MediaDescriptor& rMediaDescriptor,
AutoRecovery::TDocumentInfo& rInfo );
/** @short notifies all interested listener about the current state of the currently running operation.
@descr We support different set's of functions. Job::AutoSave, Job::EmergencySave, Job::Recovery ... etcpp. Listener can register itself for any type of supported functionality ... but not for document URL's in special.
@param eJob is used to know, which set of listener we must notify.
/** short create a feature event struct, which can be send to any interested listener.
@param eJob describe the current running operation Job::AutoSave, Job::EmergencySave, Job::Recovery
@param sEventType describe the type of this event START, STOP, UPDATE
@param pInfo if sOperation is an update, this parameter must be different from NULL and is used to send information regarding the current handled document.
// TODO document me void implts_resetHandleStates();
// TODO document me void implts_specifyDefaultFilterAndExtension(AutoRecovery::TDocumentInfo& rInfo);
// TODO document me void implts_specifyAppModuleAndFactory(AutoRecovery::TDocumentInfo& rInfo);
/** retrieves the names of all active views of the given document @param rInfo the document info, whose <code>Document</code> member must not be <NULL/>.
*/ void implts_collectActiveViewNames( AutoRecovery::TDocumentInfo& rInfo );
/** updates the configuration so that for all documents, their current view/names are stored
*/ void implts_persistAllActiveViewNames();
// TODO document me void implts_prepareEmergencySave();
// TODO document me void implts_doEmergencySave(const DispatchParams& aParams);
// TODO document me void implts_doRecovery(const DispatchParams& aParams);
// TODO document me void implts_doSessionSave(const DispatchParams& aParams);
// TODO document me void implts_doSessionQuietQuit();
// TODO document me void implts_doSessionRestore(const DispatchParams& aParams);
// TODO document me void implts_backupWorkingEntry(const DispatchParams& aParams);
// TODO document me void implts_cleanUpWorkingEntry(const DispatchParams& aParams);
/** try to make sure that all changed config items (not our used config access only) will be flushed back to disc.
E.g. our svtools::ConfigItems() has to be flushed explicitly .-(
Note: This method can't fail. Flushing of config entries is an optional feature. Errors can be ignored.
*/ staticvoid impl_flushALLConfigChanges();
// TODO document me
AutoRecovery::EFailureSafeResult implts_copyFile(const OUString& sSource , const OUString& sTargetPath, const OUString& sTargetName);
/** @short converts m_eJob into a job description, which can be used to inform an outside listener about the current running operation
@param eJob describe the current running operation Job::AutoSave, Job::EmergencySave, Job::Recovery
@return [string] a suitable job description of form: vnd.sun.star.autorecovery:/do...
*/ static OUString implst_getJobDescription(Job eJob);
/** @short map the given URL to an internal int representation.
@param aURL the url, which describe the next starting or may be already running operation.
@return [long] the internal int representation see enum class Job
*/ static Job implst_classifyJob(const css::util::URL& aURL);
/// TODO void implts_verifyCacheAgainstDesktopDocumentList();
/// TODO document me bool impl_enoughDiscSpace(sal_Int32 nRequiredSpace);
/// TODO document me staticvoid impl_showFullDiscError();
/** @short try to create/use a progress and set it inside the environment.
@descr The problem behind: There exists different use case of this method. a) An external progress is provided by our CrashSave or Recovery dialog. b) We must create our own progress e.g. for an AutoSave c) Sometimes our application filters don't use the progress provided by the MediaDescriptor. They use the Frame every time to create its own progress. So we implemented a HACK for these and now we set an InterceptedProgress there for the time WE use this frame for loading/storing documents .-)
@param xNewFrame must be set only in case WE create a new frame (e.g. for loading documents on session restore or recovery). Then search for a frame using rInfo.Document must be suppressed and xFrame must be preferred instead .-)
@param rInfo used e.g. to find the frame corresponding to a document. This frame must be used to create a new progress e.g. for an AutoSave.
@param rArgs is used to set the new created progress as parameter on these set.
*/ staticvoid impl_establishProgress(const AutoRecovery::TDocumentInfo& rInfo ,
utl::MediaDescriptor& rArgs , const css::uno::Reference< css::frame::XFrame >& xNewFrame);
Every URL supported by our UCB component can be used here. Further it doesn't matter if the file really exists or not. Because removing a non existent file will have the same result at the end... a non existing file .-)
On the other side removing of files from disc is an optional feature. If we are not able doing so... it's not a real problem. Ok - users disc place will be smaller then... but we should produce a crash during crash save because we can't delete a temporary file only!
@param sURL the url of the file, which should be removed.
*/ void st_impl_removeFile(const OUString& sURL);
/** try to remove ".lock" file from disc if office will be terminated not using the official way .-)
This method has to be handled "optional". So every error inside has to be ignored ! This method CAN NOT FAIL ... it can forget something only .-)
*/ void st_impl_removeLockFile();
};
constexpr OUString CMD_DO_AUTO_SAVE = u"/doAutoSave"_ustr; // force AutoSave ignoring the AutoSave timer
constexpr OUString CMD_DO_PREPARE_EMERGENCY_SAVE = u"/doPrepareEmergencySave"_ustr; // prepare the office for the following EmergencySave step (hide windows etcpp.)
constexpr OUString CMD_DO_EMERGENCY_SAVE = u"/doEmergencySave"_ustr; // do EmergencySave on crash
constexpr OUString CMD_DO_RECOVERY = u"/doAutoRecovery"_ustr; // recover all crashed documents
constexpr OUString CMD_DO_ENTRY_BACKUP = u"/doEntryBackup"_ustr; // try to store a temp or original file to a user defined location
constexpr OUString CMD_DO_ENTRY_CLEANUP = u"/doEntryCleanUp"_ustr; // remove the specified entry from the recovery cache
constexpr OUString CMD_DO_SESSION_SAVE = u"/doSessionSave"_ustr; // save all open documents if e.g. a window manager closes an user session
constexpr OUString CMD_DO_SESSION_QUIET_QUIT = u"/doSessionQuietQuit"_ustr; // let the current session be quietly closed ( the saving should be done using doSessionSave previously ) if e.g. a window manager closes an user session
constexpr OUString CMD_DO_SESSION_RESTORE = u"/doSessionRestore"_ustr; // restore a saved user session from disc
constexpr OUString CMD_DO_DISABLE_RECOVERY = u"/disableRecovery"_ustr; // disable recovery and auto save (!) temp. for this office session
constexpr OUString CMD_DO_SET_AUTOSAVE_STATE = u"/setAutoSaveState"_ustr; // disable/enable auto save (not crash save) for this office session
const sal_Int32 MIN_DISCSPACE_DOCSAVE = 5; // [MB] const sal_Int32 MIN_DISCSPACE_CONFIGSAVE = 1; // [MB] const sal_Int32 RETRY_STORE_ON_FULL_DISC_FOREVER = 300; // not forever ... but often enough .-) const sal_Int32 RETRY_STORE_ON_MIGHT_FULL_DISC_USEFULL = 3; // in case FULL DISC does not seem the real problem const sal_Int32 GIVE_UP_RETRY = 1; // in case FULL DISC does not seem the real problem
#define MIN_TIME_FOR_USER_IDLE 10000 // 10s user idle
// enable the following defines in case you wish to simulate a full disc for debug purposes .-)
// this define throws every time a document is stored or a configuration change // should be flushed an exception ... so the special error handler for this scenario is triggered // #define TRIGGER_FULL_DISC_CHECK
// force "return sal_False" for the method impl_enoughDiscSpace(). // #define SIMULATE_FULL_DISC
class CacheLockGuard
{ private:
// holds the outside caller alive, so it's shared resources // are valid every time
css::uno::Reference< css::uno::XInterface > m_xOwner;
// mutex shared with outside caller!
osl::Mutex& m_rSharedMutex;
// this variable knows the state of the "cache lock"
sal_Int32& m_rCacheLock;
// to prevent increasing/decreasing of m_rCacheLock more than once // we must know if THIS guard has an actual lock set there! bool m_bLockedByThisGuard;
// This cache lock is needed only to prevent us from removing/adding // items from/into the recovery cache ... during it's used at another code place // for iterating .-)
// Modifying of item properties is allowed and sometimes needed! // So we should detect only the dangerous state of concurrent add/remove // requests and throw an exception then ... which can of course break the whole // operation. On the other side a crash reasoned by an invalid stl iterator // will have the same effect .-)
if ( (m_rCacheLock > 0) && bLockForAddRemoveVectorItems )
{
OSL_FAIL("Re-entrance problem detected. Using of an stl structure in combination with iteration, adding, removing of elements etcpp."); throw css::uno::RuntimeException(
u"Re-entrance problem detected. Using of an stl structure in combination with iteration, adding, removing of elements etcpp."_ustr,
m_xOwner);
}
if (m_rCacheLock < 0)
{
OSL_FAIL("Wrong using of member m_nDocCacheLock detected. A ref counted value shouldn't reach values <0 .-)"); throw css::uno::RuntimeException(
u"Wrong using of member m_nDocCacheLock detected. A ref counted value shouldn't reach values <0 .-)"_ustr,
m_xOwner);
} /* SAFE */
}
void AutoRecovery::initListeners()
{ // read configuration to know if autosave/recovery is on/off etcpp...
implts_readConfig();
implts_startListening();
// establish callback for our internal used timer. // Note: Its only active, if the timer will be started ...
SolarMutexGuard g;
m_aTimer.SetInvokeHandler(LINK(this, AutoRecovery, implts_timerExpired));
}
// still running operation ... ignoring Job::AutoSave. // All other requests has higher prio! if (
( m_eJob != Job::NoJob ) &&
((m_eJob & Job::AutoSave ) != Job::AutoSave)
)
{
SAL_INFO("fwk.autorecovery", "AutoRecovery::dispatch(): There is already an asynchronous dispatch() running. New request will be ignored!"); return;
}
// check if somewhere wish to disable recovery temp. for this office session // This can be done immediately... must not been done asynchronous. if ((eNewJob & Job::DisableAutorecovery) == Job::DisableAutorecovery)
{ // it's important to set a flag internally, so AutoRecovery will be suppressed - even if it's requested.
m_eJob |= eNewJob;
implts_stopTimer();
implts_stopListening(); return;
}
// disable/enable AutoSave for this office session only // independent from the configuration entry. if ((eNewJob & Job::SetAutosaveState) == Job::SetAutosaveState)
{ bool bOn = lArgs.getUnpackedValueOrDefault(PROP_AUTOSAVE_STATE, true); if (bOn)
{ // don't enable AutoSave hardly ! // reload configuration to know the current state.
implts_readAutoSaveConfig();
g.clear();
implts_updateTimer(); // can it happen that might be the listener was stopped? .-) // make sure it runs always... even if AutoSave itself was disabled temporarily.
implts_startListening();
} else
{
implts_stopTimer();
m_eJob &= ~Job::AutoSave;
m_eTimerType = AutoRecovery::E_DONT_START_TIMER;
} return;
}
// in case a new dispatch overwrites a may ba active AutoSave session // we must restore this session later. see below ... bool bWasAutoSaveActive = ((eJob & Job::AutoSave) == Job::AutoSave); bool bWasUserAutoSaveActive =
((eJob & Job::UserAutoSave) == Job::UserAutoSave);
// On the other side it makes no sense to reactivate the AutoSave operation // if the new dispatch indicates a final decision... // E.g. an EmergencySave/SessionSave indicates the end of life of the current office session. // It makes no sense to reactivate an AutoSave then. // But a Recovery or SessionRestore should reactivate a may be already active AutoSave. bool bAllowAutoSaveReactivation = true;
// new document => put it into the internal list if (
(aEvent.EventName == EVENT_ON_NEW) ||
(aEvent.EventName == EVENT_ON_LOAD)
)
{
implts_registerDocument(xDocument);
} // document modified => set its modify state new (means modified against the original file!) elseif ( aEvent.EventName == EVENT_ON_MODIFYCHANGED )
{
implts_updateModifiedState(xDocument);
} /* at least one document starts saving process => Our application code is not ready for multiple save requests at the same time. So we have to suppress our AutoSave feature for the moment, till this other save requests will be finished.
*/ elseif (
(aEvent.EventName == EVENT_ON_SAVE) ||
(aEvent.EventName == EVENT_ON_SAVEAS) ||
(aEvent.EventName == EVENT_ON_SAVETO)
)
{
implts_updateDocumentUsedForSavingState(xDocument, SAVE_IN_PROGRESS);
} // document saved => remove tmp. files - but hold config entries alive! elseif (
(aEvent.EventName == EVENT_ON_SAVEDONE) ||
(aEvent.EventName == EVENT_ON_SAVEASDONE)
)
{
SolarMutexGuard g;
implts_markDocumentAsSaved(xDocument);
implts_updateDocumentUsedForSavingState(xDocument, SAVE_FINISHED);
} /* document saved as copy => mark it as "non used by concurrent save operation". so we can try to create a backup copy if next time AutoSave is started too. Don't remove temp. files or change the modified state of the document! It was not really saved to the original file...
*/ elseif ( aEvent.EventName == EVENT_ON_SAVETODONE )
{
implts_updateDocumentUsedForSavingState(xDocument, SAVE_FINISHED);
} // If saving of a document failed by an error ... we have to save this document // by ourself next time AutoSave or EmergencySave is triggered. // But we can reset the state "used for other save requests". Otherwise // these documents will never be saved! elseif (
(aEvent.EventName == EVENT_ON_SAVEFAILED) ||
(aEvent.EventName == EVENT_ON_SAVEASFAILED) ||
(aEvent.EventName == EVENT_ON_SAVETOFAILED)
)
{
implts_updateDocumentUsedForSavingState(xDocument, SAVE_FINISHED);
} // document closed => remove temp. files and configuration entries elseif ( aEvent.EventName == EVENT_ON_UNLOAD )
{
implts_deregisterDocument(xDocument); // sal_True => stop listening for disposing() !
}
}
// Changes of the configuration must be ignored if AutoSave/Recovery was disabled for this // office session. That can happen if e.g. the command line arguments "--norestore" or "--headless" // was set. if ((m_eJob & Job::DisableAutorecovery) == Job::DisableAutorecovery) return;
for (i=0; i<c; ++i)
{
OUString sPath;
pChanges[i].Accessor >>= sPath;
// Note: This call stops the timer and starts it again. // But it checks the different timer states internally and // may be suppress the restart!
implts_updateTimer();
}
if (aEvent.Source == m_xNewDocBroadcaster)
{
m_xNewDocBroadcaster.clear(); return;
}
if (aEvent.Source == m_xRecoveryCFG)
{
m_xRecoveryCFG.clear(); return;
}
// dispose from one of our cached documents ? // Normally they should send a OnUnload message ... // But some stacktraces shows another possible use case .-)
css::uno::Reference< css::frame::XModel > xDocument(aEvent.Source, css::uno::UNO_QUERY); if (xDocument.is())
{
implts_deregisterDocument(xDocument, false); // sal_False => don't call removeEventListener() .. because it's not needed here return;
}
// throws a RuntimeException if an error occurs!
css::uno::Reference<css::container::XNameAccess> xCFG(
xConfigProvider->createInstanceWithArguments(
u"com.sun.star.configuration.ConfigurationAccess"_ustr,
comphelper::containerToSequence(lParams)),
css::uno::UNO_QUERY);
try
{
nMinSpaceDocSave = officecfg::Office::Recovery::AutoSave::MinSpaceDocSave::get();
nMinSpaceConfigSave = officecfg::Office::Recovery::AutoSave::MinSpaceConfigSave::get();
} catch(const css::uno::Exception&)
{ // These config keys are not sooooo important, that // we are interested on errors here really .-)
nMinSpaceDocSave = MIN_DISCSPACE_DOCSAVE;
nMinSpaceConfigSave = MIN_DISCSPACE_CONFIGSAVE;
}
if (pItems[i].startsWith(RECOVERY_ITEM_BASE_IDENTIFIER))
{
std::u16string_view sID = pItems[i].subView(RECOVERY_ITEM_BASE_IDENTIFIER.getLength());
aInfo.ID = o3tl::toInt32(sID); /* SAFE */ {
osl::MutexGuard g(cppu::WeakComponentImplHelperBase::rBHelper.rMutex); if (aInfo.ID > m_nIdPool)
{
m_nIdPool = aInfo.ID+1;
SAL_WARN_IF(m_nIdPool<0, "fwk.autorecovery", "AutoRecovery::implts_readConfig(): Overflow of IDPool detected!");
}
} /* SAFE */
} else
SAL_INFO("fwk.autorecovery", "AutoRecovery::implts_readConfig(): Who changed numbering of recovery items? Cache will be inconsistent then! I do not know, what will happen next time .-)");
void AutoRecovery::implts_specifyDefaultFilterAndExtension(AutoRecovery::TDocumentInfo& rInfo)
{ if (rInfo.AppModule.isEmpty())
{ throw css::uno::RuntimeException(
u"Can not find out the default filter and its extension, if no application module is known!"_ustr, static_cast< css::frame::XDispatch* >(this));
}
try
{ if (! xCFG.is())
{
implts_openConfig(); // open module config on demand and cache the update access
xCFG.set(officecfg::Setup::Office::Factories::get(),
css::uno::UNO_SET_THROW);
void AutoRecovery::implts_specifyAppModuleAndFactory(AutoRecovery::TDocumentInfo& rInfo)
{
ENSURE_OR_THROW2(
!rInfo.AppModule.isEmpty() || rInfo.Document.is(), "Can not find out the application module nor its factory URL, if no application module (or a suitable) document is known!",
*this );
void AutoRecovery::implts_collectActiveViewNames( AutoRecovery::TDocumentInfo& i_rInfo )
{
ENSURE_OR_THROW2( i_rInfo.Document.is(), "need at document, at the very least", *this );
i_rInfo.ViewNames.realloc(0);
// obtain list of controllers of this document
::std::vector< OUString > aViewNames; const Reference< XModel2 > xModel( i_rInfo.Document, UNO_QUERY ); if ( xModel.is() )
{ const Reference< css::container::XEnumeration > xEnumControllers( xModel->getControllers() ); while ( xEnumControllers->hasMoreElements() )
{ const Reference< XController2 > xController( xEnumControllers->nextElement(), UNO_QUERY );
OUString sViewName; if ( xController.is() )
sViewName = xController->getViewControllerName();
OSL_ENSURE( !sViewName.isEmpty(), "AutoRecovery::implts_collectActiveViewNames: (no XController2 ->) no view name -> no recovery of this view!" );
if ( !sViewName.isEmpty() )
aViewNames.push_back( sViewName );
}
}
// This list will be filled with every document for (auto & elem : m_lDocCache)
{
implts_collectActiveViewNames(elem);
implts_flushConfigItem(elem);
}
}
// remove if (bRemoveIt)
{ // Catch NoSuchElementException. // It's not a good idea inside multithreaded environments to call hasElement - removeElement. // DO IT! try
{
osl::File::remove(rInfo.OldTempURL);
osl::File::remove(rInfo.NewTempURL);
rInfo.OldTempURL.clear();
rInfo.NewTempURL.clear();
xModify->removeByName(sID);
} catch(const css::container::NoSuchElementException&)
{ return;
}
} else
{ // new/modify
css::uno::Reference< css::beans::XPropertySet > xSet; bool bNew = !xCheck->hasByName(sID); if (bNew)
{ if (!bAllowAdd) return; // no change made, just exit
if (bNew)
xModify->insertByName(sID, css::uno::Any(xSet));
}
} catch(const css::uno::RuntimeException&)
{ throw;
} catch(const css::uno::Exception&)
{ // ??? can it happen that a full disc let these set of operations fail too ???
}
sal_Int32 nRetry = RETRY_STORE_ON_FULL_DISC_FOREVER; do
{ try
{
batch->commit();
#ifdef TRIGGER_FULL_DISC_CHECK throw css::uno::Exception("trigger full disk check"); #else// TRIGGER_FULL_DISC_CHECK
nRetry = 0; #endif// TRIGGER_FULL_DISC_CHECK
} catch(const css::uno::Exception&)
{ // a) FULL DISC seems to be the problem behind => show error and retry it forever (e.g. retry=300) // b) unknown problem (may be locking problem) => reset RETRY value to more useful value(!) (e.g. retry=3) // c) unknown problem (may be locking problem) + 1..2 repeating operations => throw the original exception to force generation of a stacktrace !
if (! impl_enoughDiscSpace(nMinSpaceConfigSave))
AutoRecovery::impl_showFullDiscError(); elseif (nRetry > RETRY_STORE_ON_MIGHT_FULL_DISC_USEFULL)
nRetry = RETRY_STORE_ON_MIGHT_FULL_DISC_USEFULL; elseif (nRetry <= GIVE_UP_RETRY) throw; // force stacktrace to know if there exist might other reasons, why an AutoSave can fail !!!
void AutoRecovery::implts_stopListening()
{
css::uno::Reference< css::util::XChangesNotifier > xCFG;
css::uno::Reference< css::document::XDocumentEventBroadcaster > xGlobalEventBroadcaster; /* SAFE */ {
osl::MutexGuard g(cppu::WeakComponentImplHelperBase::rBHelper.rMutex); // Attention: Don't reset our internal members here too. // May be we must work with our configuration, but don't wish to be informed // about changes any longer. Needed e.g. during Job::EmergencySave!
xCFG.set (m_xRecoveryCFG , css::uno::UNO_QUERY);
xGlobalEventBroadcaster = m_xNewDocBroadcaster;
} /* SAFE */
if (xGlobalEventBroadcaster.is() && m_bListenForDocEvents)
{
xGlobalEventBroadcaster->removeDocumentEventListener(m_xNewDocBroadcasterListener);
m_bListenForDocEvents = false;
}
if (
(m_eJob == Job::NoJob ) || // TODO may be superfluous - E_DONT_START_TIMER should be used only
(m_eTimerType == AutoRecovery::E_DONT_START_TIMER)
) return;
if (m_eTimerType == AutoRecovery::E_NORMAL_AUTOSAVE_INTERVALL)
{ const sal_Int64 nConfiguredAutoSaveInterval
= officecfg::Office::Recovery::AutoSave::TimeIntervall::get()
* sal_Int64(60000); // [min] => 60.000 ms
nMilliSeconds = nConfiguredAutoSaveInterval;
// Calculate how soon the nearest dirty document's autosave time is; // store the shortest document autosave timeout as the next timer timeout. for (constauto& docInfo : m_lDocCache)
{ if (auto xDocRecovery2 = docInfo.Document.query<XDocumentRecovery2>())
{
sal_Int64 nDirtyDuration = xDocRecovery2->getModifiedStateDuration(); if (nDirtyDuration < 0) continue; if (nDirtyDuration > nConfiguredAutoSaveInterval)
nDirtyDuration = nConfiguredAutoSaveInterval; // nMilliSeconds will be 0
nMilliSeconds
= std::min(nMilliSeconds, nConfiguredAutoSaveInterval - nDirtyDuration);
}
}
} elseif (m_eTimerType == AutoRecovery::E_POLL_FOR_USER_IDLE)
{
nMilliSeconds = MIN_TIME_FOR_USER_IDLE;
} elseif (m_eTimerType == AutoRecovery::E_POLL_TILL_AUTOSAVE_IS_ALLOWED)
nMilliSeconds = 300; // there is a minimum time frame, where the user can lose some key input data!
if (!m_aTimer.IsActive()) return;
m_aTimer.Stop();
}
IMPL_LINK_NOARG(AutoRecovery, implts_timerExpired, Timer *, void)
{ try
{ // This method is called by using a pointer to us. // But we must be aware that we can be destroyed hardly // if our uno reference will be gone! // => Hold this object alive till this method finish its work.
css::uno::Reference< css::uno::XInterface > xSelfHold(static_cast< css::lang::XTypeProvider* >(this));
// Needed! Otherwise every reschedule request allow a new triggered timer event :-(
implts_stopTimer();
// The timer must be ignored if AutoSave/Recovery was disabled for this // office session. That can happen if e.g. the command line arguments "--norestore" or "--headless" // was set. But normally the timer was disabled if recovery was disabled ... // But so we are more "safe" .-) /* SAFE */ {
osl::MutexGuard g(cppu::WeakComponentImplHelperBase::rBHelper.rMutex); if ((m_eJob & Job::DisableAutorecovery) == Job::DisableAutorecovery) return;
} /* SAFE */
// check some "states", where it's not allowed (better: not a good idea) to // start an AutoSave. (e.g. if the user makes drag & drop ...) // Then we poll till this "disallowed" state is gone. bool bAutoSaveNotAllowed = Application::IsUICaptured(); if (bAutoSaveNotAllowed)
{ /* SAFE */ {
osl::MutexGuard g(cppu::WeakComponentImplHelperBase::rBHelper.rMutex);
m_eTimerType = AutoRecovery::E_POLL_TILL_AUTOSAVE_IS_ALLOWED;
} /* SAFE */
implts_updateTimer(); return;
}
// analyze timer type. // If we poll for an user idle period, may be we must // do nothing here and start the timer again. /* SAFE */ {
osl::ClearableMutexGuard g(cppu::WeakComponentImplHelperBase::rBHelper.rMutex);
if (m_eTimerType == AutoRecovery::E_POLL_FOR_USER_IDLE)
{ bool bUserIdle = Application::GetLastInputInterval() > MIN_TIME_FOR_USER_IDLE; if (!bUserIdle)
{
g.clear();
implts_updateTimer(); return;
}
}
// force save of all currently open documents // The called method returns an info, if and how this // timer must be restarted. constbool bIsAlreadyIdle(m_eTimerType == AutoRecovery::E_POLL_FOR_USER_IDLE);
AutoRecovery::ETimerType eSuggestedTimer
= implts_saveDocs(/*AllowUserIdleLoop=*/!bIsAlreadyIdle, /*RemoveLockFiles=*/false);
// If timer is not used for "short callbacks" (means polling // for special states) ... reset the handle state of all // cache items. Such handle state indicates, that a document // was already saved during the THIS(!) AutoSave session. // Of course NEXT AutoSave session must be started without // any "handle" state ... if (
(eSuggestedTimer == AutoRecovery::E_DONT_START_TIMER ) ||
(eSuggestedTimer == AutoRecovery::E_NORMAL_AUTOSAVE_INTERVALL)
)
{
implts_resetHandleStates();
}
void AutoRecovery::implts_registerDocument(const css::uno::Reference< css::frame::XModel3 > & xDocument)
{ // ignore corrupted events, where no document is given ... Runtime Error ?! if (!xDocument.is()) return;
// notification for already existing document ! // Can happen if events came in asynchronous on recovery time. // Then our cache was filled from the configuration ... but now we get some // asynchronous events from the global event broadcaster. We must be sure that // we don't add the same document more than once.
AutoRecovery::TDocumentList::iterator pIt = AutoRecovery::impl_searchDocument(m_lDocCache, xDocument); if (pIt != m_lDocCache.end())
{ // Normally nothing must be done for this "late" notification. // But may be the modified state was changed in between. // Check it...
implts_updateModifiedState(xDocument); return;
}
// check if this document must be ignored for recovery ! // Some use cases don't wish support for AutoSave/Recovery ... as e.g. OLE-Server / ActiveX Control etcpp. bool bNoAutoSave = lDescriptor.getUnpackedValueOrDefault(utl::MediaDescriptor::PROP_NOAUTOSAVE, false); if (bNoAutoSave) return;
// Check if doc is well known on the desktop. Otherwise ignore it! // Other frames mostly are used from external programs - e.g. the bean ...
css::uno::Reference< css::frame::XController > xController = xDocument->getCurrentController(); if (!xController.is()) return;
css::uno::Reference< css::frame::XFrame > xFrame = xController->getFrame(); if (!xFrame.is()) return;
css::uno::Reference< css::frame::XDesktop > xDesktop (xFrame->getCreator(), css::uno::UNO_QUERY); if (!xDesktop.is()) return;
// if the document doesn't support the XDocumentRecovery interface, we're not interested in it.
Reference< XDocumentRecovery > xDocRecovery( xDocument, UNO_QUERY ); if ( !xDocRecovery.is() ) return;
// get all needed information of this document // We need it to update our cache or to locate already existing elements there!
AutoRecovery::TDocumentInfo aNew;
aNew.Document = xDocument;
// TODO replace getLocation() with getURL() ... it's a workaround currently only!
css::uno::Reference< css::frame::XStorable > xDoc(aNew.Document, css::uno::UNO_QUERY_THROW);
aNew.OrgURL = xDoc->getLocation();
// classify the used application module, which is used by this document.
implts_specifyAppModuleAndFactory(aNew);
// Hack! Check for "illegal office documents"... as e.g. the Basic IDE // It's not really a full featured office document. It doesn't provide a URL, any filter, a factory URL etcpp. // TODO file bug to Basic IDE developers. They must remove the office document API from its service. if (
(aNew.OrgURL.isEmpty()) &&
(aNew.FactoryURL.isEmpty())
)
{
OSL_FAIL( "AutoRecovery::implts_registerDocument: this should not happen anymore!" ); // nowadays, the Basic IDE should already die on the "supports XDocumentRecovery" check. And no other known // document type fits in here ... return;
}
// By the way - get some information about the default format for saving! // and save an information about the real used filter by this document. // We save this document with DefaultFilter ... and load it with the RealFilter.
implts_specifyDefaultFilterAndExtension(aNew);
aNew.RealFilter = lDescriptor.getUnpackedValueOrDefault(utl::MediaDescriptor::PROP_FILTERNAME, OUString());
// Further we must know, if this document base on a template. // Then we must load it in a different way.
css::uno::Reference< css::document::XDocumentPropertiesSupplier > xSupplier(aNew.Document, css::uno::UNO_QUERY); if (xSupplier.is()) // optional interface!
{
css::uno::Reference< css::document::XDocumentProperties > xDocProps(xSupplier->getDocumentProperties(), css::uno::UNO_SET_THROW);
aNew.TemplateURL = xDocProps->getTemplateURL();
}
// create a new cache entry ... this document is not known.
++m_nIdPool;
aNew.ID = m_nIdPool;
SAL_WARN_IF(m_nIdPool<0, "fwk.autorecovery", "AutoRecovery::implts_registerDocument(): Overflow of ID pool detected.");
m_lDocCache.push_back(aNew);
// Don't register new documents here. (They are registered elsewhere when saved or modified...)
implts_flushConfigItem(aInfo, /*bRemoveIt=*/false, /*bAllowAdd=*/!aNew.OrgURL.isEmpty());
implts_startModifyListeningOnDoc(aInfo);
// Attention: Don't leave SAFE section, if you work with pIt! // Because it points directly into the m_lDocCache list ...
CacheLockGuard aCacheLock(this, cppu::WeakComponentImplHelperBase::rBHelper.rMutex, m_nDocCacheLock, LOCK_FOR_CACHE_USE);
AutoRecovery::TDocumentList::iterator pIt = AutoRecovery::impl_searchDocument(m_lDocCache, xDocument); if (pIt == m_lDocCache.end()) return; // unknown document => not a runtime error! Because we register only a few documents. see registration ...
aInfo = *pIt;
aCacheLock.unlock();
// Sometimes we close documents by ourself. // And these documents can't be deregistered. // Otherwise we lose our configuration data... but need it ! // see SessionSave ! if (aInfo.IgnoreClosing) return;
CacheLockGuard aCacheLock2(this, cppu::WeakComponentImplHelperBase::rBHelper.rMutex, m_nDocCacheLock, LOCK_FOR_CACHE_ADD_REMOVE);
pIt = AutoRecovery::impl_searchDocument(m_lDocCache, xDocument); if (pIt != m_lDocCache.end())
m_lDocCache.erase(pIt);
pIt = m_lDocCache.end(); // otherwise it's not specified what pIt means!
aCacheLock2.unlock();
} /* SAFE */
/* This method is called within disposing() of the document too. But there it's not a good idea to deregister us as listener. Further it makes no sense - because the broadcaster dies. So we suppress deregistration in such case...
*/ if (bStopListening)
implts_stopModifyListeningOnDoc(aInfo);
implts_flushConfigItem(aInfo, true); // sal_True => remove it from config
}
AutoRecovery::TDocumentList::iterator pIt = AutoRecovery::impl_searchDocument(m_lDocCache, xDocument); if (pIt != m_lDocCache.end())
{ /* Now we know, that this document was modified again and must be saved next time. But we don't need this information for every e.g. key input of the user. So we stop listening here. But if the document was saved as temp. file we start listening for this event again.
*/
implts_stopModifyListeningOnDoc(*pIt);
}
} /* SAFE */
}
void AutoRecovery::implts_updateModifiedState(const css::uno::Reference< css::frame::XModel >& xDocument)
{ // use true as fallback to get every document on EmergencySave/AutoRecovery! bool bModified = true;
css::uno::Reference< css::util::XModifiable > xModify(xDocument, css::uno::UNO_QUERY); if (xModify.is())
bModified = xModify->isModified();
/* Since the document has been saved, update its entry in the document * cache. We essentially reset the state of the document from an * autorecovery perspective, updating things like the filename (which * would change in the case of a 'Save as' operation) and the associated
* backup file URL. */
aInfo.DocumentState = DocState::Unknown; // TODO replace getLocation() with getURL() ... it's a workaround currently only!
css::uno::Reference< css::frame::XStorable > xDoc(aInfo.Document, css::uno::UNO_QUERY);
aInfo.OrgURL = xDoc->getLocation();
/* Save off the backup file URLs and then clear them. NOTE - it is * important that we clear them - otherwise, we could enter a state * where pIt->OldTempURL == pIt->NewTempURL and our backup algorithm * in implts_saveOneDoc will write to that URL and then delete the file
* at that URL (bug #96607) */
sRemoveURL1 = aInfo.OldTempURL;
sRemoveURL2 = aInfo.NewTempURL;
aInfo.OldTempURL.clear();
aInfo.NewTempURL.clear();
/* Currently the document is not closed in case of crash, so the lock file must be removed explicitly
*/ void lc_removeLockFile(AutoRecovery::TDocumentInfo const & rInfo)
{ #if !HAVE_FEATURE_MULTIUSER_ENVIRONMENT || HAVE_FEATURE_MACOSX_SANDBOX
(void) rInfo; #else if ( !rInfo.Document.is() ) return;
for (auto & info : m_lDocCache)
{ // WORKAROUND... Since the documents are not closed the lock file must be removed explicitly // it is not done on documents saving since shutdown can be cancelled
lc_removeLockFile( info );
// Prevent us from deregistration of these documents. // Because we close these documents by ourself (see XClosable below) ... // it's fact, that we reach our deregistration method. There we // must not(!) update our configuration ... Otherwise all // session data are lost !!!
info.IgnoreClosing = true;
// reset modified flag of these documents (ignoring the notification about it!) // Otherwise a message box is shown on closing these models.
implts_stopModifyListeningOnDoc(info);
// if the session save is still running the documents should not be thrown away, // actually that would be a bad sign, that means that the SessionManager tries // to kill the session before the saving is ready if ((m_eJob & Job::SessionSave) != Job::SessionSave)
{
css::uno::Reference< css::util::XModifiable > xModify(info.Document, css::uno::UNO_QUERY); if (xModify.is())
xModify->setModified(false);
// close the model.
css::uno::Reference< css::util::XCloseable > xClose(info.Document, css::uno::UNO_QUERY); if (xClose.is())
{ try
{
xClose->close(false);
} catch(const css::uno::Exception&)
{ // At least it's only a try to close these documents before anybody else it does. // So it seems to be possible to ignore any error here .-)
}
info.Document.clear();
}
}
}
aCacheLock.unlock();
} /* SAFE */
}
/* TODO WORKAROUND:
#i64599#
Normally the MediaDescriptor argument NoAutoSave indicates, that a document must be ignored for AutoSave and Recovery. But sometimes XModel->getArgs() does not contained this information if implts_registerDocument() was called. So we have to check a second time, if this property is set... Best place doing so is to check it immediately before saving and suppressing saving the document then. Of course removing the corresponding cache entry is not an option. Because it would disturb iteration over the cache! So we ignore such documents only... Hopefully next time they are not inserted in our cache.
*/ bool lc_checkIfSaveForbiddenByArguments(AutoRecovery::TDocumentInfo const & rInfo)
{ if (! rInfo.Document.is()) returntrue;
// Set the default timer action for our call. // Default = NORMAL_AUTOSAVE // We return a suggestion for an active timer only. // It will be ignored if the timer was disabled by the user ... // Further this state can be set to USER_IDLE only later in this method. // It's not allowed to reset such state then. Because we must know, if // there exists POSTPONED documents. see below ...
AutoRecovery::ETimerType eTimer = AutoRecovery::E_NORMAL_AUTOSAVE_INTERVALL;
// This list will be filled with every document // which should be saved as last one. E.g. if it was used // already for a UI save operation => crashed ... and // now we try to save it again ... which can fail again ( of course .-) ).
::std::vector< AutoRecovery::TDocumentList::iterator > lDangerousDocs;
// WORKAROUND... Since the documents are not closed the lock file must be removed explicitly if ( bRemoveLockFiles )
lc_removeLockFile( aInfo );
// WORKAROUND ... see comment of this method if (lc_checkIfSaveForbiddenByArguments(aInfo)) continue;
// already auto saved during this session :-) // This state must be reset for all documents // if timer is started with normal AutoSaveTimerIntervall! if ((aInfo.DocumentState & DocState::Handled) == DocState::Handled) continue;
// don't allow implts_deregisterDocument to remove from RecoveryList during shutdown jobs if (m_eJob & (Job::EmergencySave | Job::SessionSave))
aInfo.IgnoreClosing = true;
// Not modified documents are not saved. // We save information about the URL only!
Reference< XDocumentRecovery > xDocRecover( aInfo.Document, UNO_QUERY_THROW ); if ( !xDocRecover->wasModifiedSinceLastSave() )
{
aInfo.DocumentState |= DocState::Handled;
*pIt = aInfo; continue;
}
// If the document became modified not too long ago, don't autosave it yet. if (bAllowUserIdleLoop)
{ if (auto xDocRecovery2 = xDocRecover.query<XDocumentRecovery2>())
{ const sal_Int64 nDirtyDuration = xDocRecovery2->getModifiedStateDuration(); // Round up to second - if this document is almost ready for autosave, do it now. if (nDirtyDuration + 999 < nConfiguredAutoSaveInterval)
{
aInfo.DocumentState |= DocState::Handled; continue;
}
}
}
// check if this document is still used by a concurrent save operation // e.g. if the user tried to save via UI. // Handle it in the following way: // i) For an AutoSave ... ignore this document! It will be saved and next time we will (hopefully) // get a notification about the state of this operation. // And if a document was saved by the user we can remove our temp. file. But that will be done inside // our callback for SaveDone notification. // ii) For a CrashSave ... add it to the list of dangerous documents and // save it after all other documents was saved successfully. That decrease // the chance for a crash inside a crash. // On the other side it's not necessary for documents, which are not modified. // They can be handled normally - means we patch the corresponding configuration entry only. // iii) For a SessionSave ... ignore it! There is no time to wait for this save operation. // Because the WindowManager will kill the process if it doesn't react immediately. // On the other side we can't risk a concurrent save request ... because we know // that it will produce a crash.
// Attention: Because eJob is used as a flag field, you have to check for the worst case first. // E.g. a CrashSave can overwrite an AutoSave. So you have to check for a CrashSave before an AutoSave! if (aInfo.UsedForSaving)
{ if ((eJob & Job::EmergencySave) == Job::EmergencySave)
{
lDangerousDocs.push_back(pIt); continue;
} else if ((eJob & Job::SessionSave) == Job::SessionSave)
{ continue;
} else if ((eJob & Job::AutoSave) == Job::AutoSave)
{
eTimer = AutoRecovery::E_POLL_TILL_AUTOSAVE_IS_ALLOWED;
aInfo.DocumentState |= DocState::Postponed; continue;
}
}
// a) Document was not postponed => wait for user idle if not urgent // b) Document was postponed => save it (because user idle/call_back was checked already) if (!(aInfo.DocumentState & DocState::Postponed))
{
aInfo.DocumentState |= DocState::Postponed;
*pIt = aInfo; // postponed documents will be saved if this method is called again! // That can be done by an outside started timer => E_POLL_FOR_USER_IDLE (if normal AutoSave is active) // or it must be done directly without starting any timer => E_CALL_ME_BACK (if Emergency- or SessionSave is active and must be finished ASAP!) if (!bAllowUserIdleLoop)
eTimer = AutoRecovery::E_CALL_ME_BACK; else
eTimer = AutoRecovery::E_POLL_FOR_USER_IDLE; continue;
}
// } /* SAFE */
g.clear(); // changing of aInfo and flushing it is done inside implts_saveOneDoc!
implts_saveOneDoc(sBackupPath, aInfo, xExternalProgress);
implts_informListener(eJob, AutoRecovery::implst_createFeatureStateEvent(eJob, OPERATION_UPDATE, &aInfo));
g.reset(); // /* SAFE */ {
*pIt = std::move(aInfo);
}
// Did we have some "dangerous candidates" ? // Try to save it ... but may be it will fail ! for (autoconst& dangerousDoc : lDangerousDocs)
{
pIt = dangerousDoc;
AutoRecovery::TDocumentInfo aInfo = *pIt;
// } /* SAFE */
g.clear(); // changing of aInfo and flushing it is done inside implts_saveOneDoc!
implts_saveOneDoc(sBackupPath, aInfo, xExternalProgress);
implts_informListener(eJob, AutoRecovery::implst_createFeatureStateEvent(eJob, OPERATION_UPDATE, &aInfo));
g.reset(); // /* SAFE */ {
*pIt = std::move(aInfo);
}
} /* SAFE */
return eTimer;
}
void AutoRecovery::implts_saveOneDoc(const OUString& sBackupPath ,
AutoRecovery::TDocumentInfo& rInfo , const css::uno::Reference< css::task::XStatusIndicator >& xExternalProgress)
{ // no document? => can occur if we loaded our configuration with files, // which couldn't be recovered successfully. In such case we have all needed information // excepting the real document instance!
// TODO: search right place, where such "dead files" can be removed from the configuration! if (!rInfo.Document.is()) return;
// if the document was loaded with a password, it should be // stored with password
utl::MediaDescriptor lNewArgs;
css::uno::Sequence< css::beans::NamedValue > aEncryptionData =
lOldArgs.getUnpackedValueOrDefault(utl::MediaDescriptor::PROP_ENCRYPTIONDATA,
css::uno::Sequence< css::beans::NamedValue >()); if (aEncryptionData.hasElements())
lNewArgs[utl::MediaDescriptor::PROP_ENCRYPTIONDATA] <<= aEncryptionData;
// Further it must be saved using the default file format of that application. // Otherwise we will some data lost. if (!rInfo.DefaultFilter.isEmpty())
lNewArgs[utl::MediaDescriptor::PROP_FILTERNAME] <<= rInfo.DefaultFilter;
// prepare frame/document/mediadescriptor in a way, that it uses OUR progress .-) if (xExternalProgress.is())
lNewArgs[utl::MediaDescriptor::PROP_STATUSINDICATOR] <<= xExternalProgress;
impl_establishProgress(rInfo, lNewArgs, css::uno::Reference< css::frame::XFrame >());
// #i66598# use special handling of property "DocumentBaseURL" (it must be an empty string!) // for make hyperlinks working
lNewArgs[utl::MediaDescriptor::PROP_DOCUMENTBASEURL] <<= OUString();
// try to save this document as a new temp file every time. // Mark AutoSave state as "INCOMPLETE" if it failed. // Because the last temp file is too old and does not include all changes.
Reference< XDocumentRecovery > xDocRecover(rInfo.Document, css::uno::UNO_QUERY_THROW);
// save the state about "trying to save" // ... we need it for recovery if e.g. a crash occurs inside next line!
rInfo.DocumentState |= DocState::TrySave; // just update existing info: don't add any recovery record until recovery file created.
implts_flushConfigItem(rInfo, /*bRemoveIt=*/false, /*bAllowAdd=*/false);
// If userautosave is enabled, first try to save the original file. // Note that we must do it *before* calling storeToRecoveryFile, so in case of failure here // we won't remain with the modified flag set to true, even though the autorecovery save succeeded. constbool bEmergencySave(m_eJob & Job::EmergencySave); bool bUserAutoSaved = false; try
{ // We must check here for an empty URL to avoid a "This operation is not supported on this operating system." // message during autosave. if (!bEmergencySave && m_eJob & Job::UserAutoSave && !rInfo.OrgURL.isEmpty())
{
Reference< XStorable > xDocSave(rInfo.Document, css::uno::UNO_QUERY_THROW);
xDocSave->store();
bUserAutoSaved = true;
}
} catch(const css::uno::Exception&)
{
}
// DocState::Modified status cannot be trusted to be accurate, but at least attempt to be so, // since this rInfo will eventually get assigned to m_lDocCache as the authoritative status. const Reference<css::util::XModifiable> xModify(rInfo.Document, UNO_QUERY); constbool bModified = xModify.is() && xModify->isModified(); if (bModified)
rInfo.DocumentState |= DocState::Modified; elseif (xModify.is())
rInfo.DocumentState &= ~DocState::Modified;
// If it is no longer modified, it is the same as on disk, and can be removed from RecoveryList. constbool bRemoveIt
= xModify.is() && !xModify->isModified() && bUserAutoSaved && !(m_eJob & Job::SessionSave);
sal_Int32 nRetry = RETRY_STORE_ON_FULL_DISC_FOREVER; bool bError = false; do
{ try
{ // skip recovery if it was already saved in-place. if (!bRemoveIt)
xDocRecover->storeToRecoveryFile(rInfo.NewTempURL,
lNewArgs.getAsConstPropertyValueList());
#ifdef TRIGGER_FULL_DISC_CHECK throw css::uno::Exception("trigger full disk check"); #else// TRIGGER_FULL_DISC_CHECK
// skip saving XLSX with protected sheets, if their passwords haven't supported yet if ( rException.Message.startsWith("SfxBaseModel::impl_store") )
{ const css::task::ErrorCodeIOException& pErrorCodeIOException = static_cast<const css::task::ErrorCodeIOException&>(rException); if ( static_cast<ErrCode>(pErrorCodeIOException.ErrCode) == ERRCODE_SFX_WRONGPASSWORD )
{ // stop and remove the bad temporary file, instead of filling the disk with them
bError = false; break;
}
}
// a) FULL DISC seems to be the problem behind => show error and retry it forever (e.g. retry=300) // b) unknown problem (may be locking problem) => reset RETRY value to more useful value(!) (e.g. retry=3) // c) unknown problem (may be locking problem) + 1..2 repeating operations => throw the original exception to force generation of a stacktrace !
if (! impl_enoughDiscSpace(nMinSpaceDocSave))
AutoRecovery::impl_showFullDiscError(); elseif (nRetry > RETRY_STORE_ON_MIGHT_FULL_DISC_USEFULL)
nRetry = RETRY_STORE_ON_MIGHT_FULL_DISC_USEFULL; elseif (nRetry <= GIVE_UP_RETRY)
{ // delete the empty file created by implts_generateNewTempURL if (tools::isEmptyFileUrl(rInfo.NewTempURL))
osl::File::remove(rInfo.NewTempURL);
throw; // force stacktrace to know if there exist might other reasons, why an AutoSave can fail !!!
}
--nRetry;
}
} while(nRetry>0);
if (! bError)
{ // safe the state about success // ... you know the reason: to know it on recovery time if next line crash .-)
rInfo.DocumentState &= ~DocState::TrySave;
rInfo.DocumentState |= DocState::Handled;
rInfo.DocumentState |= DocState::Succeeded;
} else
{ // safe the state about error ...
rInfo.NewTempURL.clear();
rInfo.DocumentState &= ~DocState::TrySave;
rInfo.DocumentState |= DocState::Handled;
rInfo.DocumentState |= DocState::Incomplete;
}
// make sure the progress is not referred any longer
impl_forgetProgress(rInfo, lNewArgs, css::uno::Reference< css::frame::XFrame >());
// try to remove the old temp file. // Ignore any error here. We have a new temp file, which is up to date. // The only thing is: we fill the disk with temp files, if we can't remove old ones :-)
OUString sRemoveFile = rInfo.OldTempURL;
rInfo.OldTempURL = rInfo.NewTempURL;
rInfo.NewTempURL.clear();
// If it is modified, a recovery file has just been created, so add to RecoveryList.
implts_flushConfigItem(rInfo, /*bRemoveIt=*/false, /*bAllowAdd=*/bModified);
// We must know if the user modifies the document again ...
implts_startModifyListeningOnDoc(rInfo);
Job eJob = m_eJob; for (auto & info : m_lDocCache)
{ // Such documents are already loaded by the last loop. // Don't check DocState::Succeeded here! It may be the final state of an AutoSave // operation before!!! if ((info.DocumentState & DocState::Handled) == DocState::Handled) continue;
// a1,b1,c1,d2,e2,f2) if ((info.DocumentState & DocState::Damaged) == DocState::Damaged)
{ // don't forget to inform listener! May be this document was // damaged on last saving time ... // Then our listener need this notification. // If it was damaged during last "try to open" ... // it will be notified more than once. SH.. HAPPENS ... // } /* SAFE */
g.clear();
implts_informListener(eJob,
AutoRecovery::implst_createFeatureStateEvent(eJob, OPERATION_UPDATE, &info));
g.reset(); // /* SAFE */ { continue;
}
utl::MediaDescriptor lDescriptor;
// it's a UI feature - so the "USER" itself must be set as referrer
lDescriptor[utl::MediaDescriptor::PROP_REFERRER] <<= OUString(REFERRER_USER);
lDescriptor[utl::MediaDescriptor::PROP_SALVAGEDFILE] <<= OUString();
// recovered documents are loaded hidden, and shown all at once, later
lDescriptor[utl::MediaDescriptor::PROP_HIDDEN] <<= true;
if (aParams.m_xProgress.is())
lDescriptor[utl::MediaDescriptor::PROP_STATUSINDICATOR] <<= aParams.m_xProgress;
bool bBackupWasTried = (
((info.DocumentState & DocState::TryLoadBackup ) == DocState::TryLoadBackup) || // temp. state!
((info.DocumentState & DocState::Incomplete ) == DocState::Incomplete ) // transport DocState::TryLoadBackup from last loop to this new one!
); bool bOriginalWasTried = ((info.DocumentState & DocState::TryLoadOriginal) == DocState::TryLoadOriginal);
if (bBackupWasTried)
{ if (!bOriginalWasTried)
{
info.DocumentState |= DocState::Incomplete; // try original URL ... ! don't continue with next item here ...
} else
{
info.DocumentState |= DocState::Damaged; continue;
}
}
// A "Salvaged" item must exists every time. The core can make something special then for recovery. // Of course it should be the real file name of the original file, in case we load the temp. backup here.
OUString sURL; if (!sLoadBackupURL.isEmpty())
{
sURL = sLoadBackupURL;
info.DocumentState |= DocState::TryLoadBackup;
lDescriptor[utl::MediaDescriptor::PROP_SALVAGEDFILE] <<= sLoadOriginalURL;
} elseif (!sLoadOriginalURL.isEmpty())
{
sURL = sLoadOriginalURL;
info.DocumentState |= DocState::TryLoadOriginal;
} else continue; // TODO ERROR!
// /* SAFE */ { // Needed for next loop!
g.reset(); continue;
}
if (!info.RealFilter.isEmpty())
{
utl::MediaDescriptor lPatchDescriptor(info.Document->getArgs());
lPatchDescriptor[utl::MediaDescriptor::PROP_FILTERNAME] <<= info.RealFilter;
info.Document->attachResource(info.Document->getURL(), lPatchDescriptor.getAsConstPropertyValueList()); // do *not* use sURL here. In case this points to the recovery file, it has already been passed // to recoverFromFile. Also, passing it here is logically wrong, as attachResource is intended // to take the logical file URL.
}
/* Normally we listen as XModifyListener on a document to know if a document was changed since our last AutoSave. And we deregister us in case we know this state. But directly after one document as recovered ... we must start listening. Otherwise the first "modify" doesn't reach us. Because we ourself called setModified() on the document via API. And currently we don't listen for any events (not at theGlobalEventBroadcaster nor at any document!).
*/
implts_startModifyListeningOnDoc(info);
// /* SAFE */ { // Needed for next loop. Don't unlock it again!
g.reset();
}
::std::vector< Reference< XComponent > > aCleanup; try
{ // create a new document of the desired type
Reference< XModel2 > xModel(m_xContext->getServiceManager()->createInstanceWithContext(
rInfo.FactoryService, m_xContext), UNO_QUERY_THROW);
aCleanup.emplace_back(xModel.get() );
// put the filter name into the descriptor - we're not going to involve any type detection, so // the document might be lost without the FilterName property if ( (rInfo.DocumentState & DocState::TryLoadOriginal) == DocState::TryLoadOriginal)
lDescriptor[ utl::MediaDescriptor::PROP_FILTERNAME ] <<= rInfo.RealFilter; else
lDescriptor[ utl::MediaDescriptor::PROP_FILTERNAME ] <<= rInfo.DefaultFilter;
if ( sURL == rInfo.FactoryURL )
{ // if the document was a new, unmodified document, then there's nothing to recover, just to init
ENSURE_OR_THROW( ( rInfo.DocumentState & DocState::Modified ) == DocState(0), "unexpected document state" );
Reference< XLoadable > xModelLoad( xModel, UNO_QUERY_THROW );
xModelLoad->initNew();
// TODO: remove load-process specific arguments from the descriptor, e.g. the status indicator
xModel->attachResource( sURL, lDescriptor.getAsConstPropertyValueList() );
} elseif (!utl::UCBContentHelper::Exists(sURL)) throw css::uno::Exception(); else
{
OUString sFilterName;
lDescriptor[utl::MediaDescriptor::PROP_FILTERNAME] >>= sFilterName; if (!sFilterName.isEmpty()
&& ( sFilterName == "Calc MS Excel 2007 XML"
|| sFilterName == "Impress MS PowerPoint 2007 XML"
|| sFilterName == "MS Word 2007 XML")) // TODO: Probably need to check other affected formats + templates?
{ // tdf#129096: in case of recovery of password protected OOXML document it is done not // the same way as ordinal loading. Inside XDocumentRecovery::recoverFromFile // there is a call to XFilter::filter which has constant media descriptor and thus // all encryption data used in document is lost. To avoid this try to walkaround // with explicit call to FormatDetector. It will try to load document, prompt for password // and store this info in media descriptor we will use for recoverFromFile call.
Reference< css::document::XExtendedFilterDetection > xDetection(
m_xContext->getServiceManager()->createInstanceWithContext(
u"com.sun.star.comp.oox.FormatDetector"_ustr, m_xContext),
UNO_QUERY_THROW);
lDescriptor[utl::MediaDescriptor::PROP_URL] <<= sURL;
Sequence< css::beans::PropertyValue > aDescriptorSeq = lDescriptor.getAsConstPropertyValueList();
OUString sType = xDetection->detect(aDescriptorSeq);
OUString sNewFilterName;
lDescriptor[utl::MediaDescriptor::PROP_FILTERNAME] >>= sNewFilterName; if (!sType.isEmpty() && sNewFilterName == sFilterName)
{ // Filter detection was okay, update media descriptor with one received from FilterDetect
lDescriptor = aDescriptorSeq;
}
}
// let it recover itself
Reference< XDocumentRecovery > xDocRecover( xModel, UNO_QUERY_THROW );
xDocRecover->recoverFromFile(
sURL,
lDescriptor.getUnpackedValueOrDefault( utl::MediaDescriptor::PROP_SALVAGEDFILE, OUString() ),
lDescriptor.getAsConstPropertyValueList()
);
// No attachResource needed here. By definition (of XDocumentRecovery), the implementation is responsible // for completely initializing the model, which includes attachResource (or equivalent), if required.
}
// re-create all the views
::std::vector< OUString > aViewsToRestore( std::cbegin(rInfo.ViewNames), std::cend(rInfo.ViewNames) ); // if we don't have views for whatever reason, then create a default-view, at least if ( aViewsToRestore.empty() )
aViewsToRestore.emplace_back( );
void AutoRecovery::implts_generateNewTempURL(const OUString& sBackupPath ,
utl::MediaDescriptor& /*rMediaDescriptor*/,
AutoRecovery::TDocumentInfo& rInfo )
{ // specify URL for saving (which points to a temp file inside backup directory) // and define a unique name, so we can locate it later. // This unique name must solve an optimization problem too! // In case we are asked to save unmodified documents too - and one of them // is an empty one (because it was new created using e.g. a URL private:factory/...) // we should not save it really. Then we put the information about such "empty document" // into the configuration and don't create any recovery file on disk. // We use the title of the document to make it unique.
OUStringBuffer sUniqueName; if (!rInfo.OrgURL.isEmpty())
{
css::uno::Reference< css::util::XURLTransformer > xParser(css::util::URLTransformer::create(m_xContext));
css::util::URL aURL;
aURL.Complete = rInfo.OrgURL;
xParser->parseStrict(aURL);
sUniqueName.append(aURL.Name);
} elseif (!rInfo.FactoryURL.isEmpty())
sUniqueName.append("untitled");
sUniqueName.append("_");
// TODO: Must we strip some illegal signs - if we use the title?
OUString AutoRecovery::implst_getJobDescription(Job eJob)
{ // describe the current running operation
OUStringBuffer sFeature(256);
sFeature.append(CMD_PROTOCOL);
// Attention: Because "eJob" is used as a flag field the order of checking these // flags is important. We must prefer job with higher priorities! // E.g. EmergencySave has an higher prio then AutoSave ... // On the other side there exist a well defined order between two different jobs. // e.g. PrepareEmergencySave must be done before EmergencySave is started of course.
void AutoRecovery::implts_prepareEmergencySave()
{ // Be sure to know all open documents really .-)
implts_verifyCacheAgainstDesktopDocumentList();
// hide all docs, so the user can't disturb our emergency save .-)
implts_changeAllDocVisibility(false);
}
void AutoRecovery::implts_doEmergencySave(const DispatchParams& aParams)
{ // Write a hint "we crashed" into the configuration, so // the error report tool is started too in case no recovery // documents exists and was saved.
// for all docs, store their current view/names in the configuration
implts_persistAllActiveViewNames();
// The called method for saving documents runs // during normal AutoSave more than once. Because // it postpone active documents and save it later. // That is normally done by recalling it from a timer. // Here we must do it immediately! // Of course this method returns the right state - // because it knows, that we are running in EMERGENCY SAVE mode .-)
boolconst bAllowUserIdleLoop = false; // not allowed to change that .-)
AutoRecovery::ETimerType eSuggestedTimer = AutoRecovery::E_DONT_START_TIMER; do
{
eSuggestedTimer = implts_saveDocs(bAllowUserIdleLoop, true, &aParams);
} while(eSuggestedTimer == AutoRecovery::E_CALL_ME_BACK);
// reset the handle state of all // cache items. Such handle state indicates, that a document // was already saved during the THIS(!) EmergencySave session. // Of course following recovery session must be started without // any "handle" state ...
implts_resetHandleStates();
// flush config cached back to disc.
impl_flushALLConfigChanges();
// try to make sure next time office will be started user won't be // notified about any other might be running office instance // remove ".lock" file from disc !
AutoRecovery::st_impl_removeLockFile();
}
// reset the handle state of all // cache items. Such handle state indicates, that a document // was already saved during the THIS(!) Recovery session. // Of course a may be following EmergencySave session must be started without // any "handle" state...
implts_resetHandleStates();
// Reset the configuration hint "we were crashed"!
std::shared_ptr<comphelper::ConfigurationChanges> batch(
comphelper::ConfigurationChanges::create());
officecfg::Office::Recovery::RecoveryInfo::Crashed::set(false, batch);
batch->commit();
}
// Be sure to know all open documents really .-)
implts_verifyCacheAgainstDesktopDocumentList();
// for all docs, store their current view/names in the configuration
implts_persistAllActiveViewNames();
// The called method for saving documents runs // during normal AutoSave more than once. Because // it postpone active documents and save it later. // That is normally done by recalling it from a timer. // Here we must do it immediately! // Of course this method returns the right state - // because it knows, that we are running in SESSION SAVE mode .-)
boolconst bAllowUserIdleLoop = false; // not allowed to change that .-)
AutoRecovery::ETimerType eSuggestedTimer = AutoRecovery::E_DONT_START_TIMER; do
{ // do not remove lock files of the documents, it will be done on session quit
eSuggestedTimer = implts_saveDocs(bAllowUserIdleLoop, false, &aParams);
} while(eSuggestedTimer == AutoRecovery::E_CALL_ME_BACK);
// reset the handle state of all // cache items. Such handle state indicates, that a document // was already saved during the THIS(!) save session. // Of course following restore session must be started without // any "handle" state ...
implts_resetHandleStates();
// flush config cached back to disc.
impl_flushALLConfigChanges();
}
// try to make sure next time office will be started user won't be // notified about any other might be running office instance // remove ".lock" file from disc! // it is done as a first action for session save since Gnome sessions // do not provide enough time for shutdown, and the dialog looks to be // confusing for the user
AutoRecovery::st_impl_removeLockFile();
// reset all modified documents, so the don't show any UI on closing ... // and close all documents, so we can shutdown the OS!
implts_prepareSessionShutdown();
// Write a hint for "stored session data" into the configuration, so // the on next startup we know what's happen last time
std::shared_ptr<comphelper::ConfigurationChanges> batch(
comphelper::ConfigurationChanges::create());
officecfg::Office::Recovery::RecoveryInfo::SessionData::set(true, batch);
batch->commit();
// flush config cached back to disc.
impl_flushALLConfigChanges();
}
// reset the handle state of all // cache items. Such handle state indicates, that a document // was already saved during the THIS(!) Restore session. // Of course a may be following save session must be started without // any "handle" state ...
implts_resetHandleStates();
// make all opened documents visible
implts_changeAllDocVisibility(true);
// Reset the configuration hint for "session save"!
SAL_INFO("fwk.autorecovery", "... reset config key 'SessionData'");
std::shared_ptr<comphelper::ConfigurationChanges> batch(
comphelper::ConfigurationChanges::create());
officecfg::Office::Recovery::RecoveryInfo::SessionData::set(false, batch);
batch->commit();
for (autoconst& info : m_lDocCache)
{ if (info.ID != aParams.m_nWorkingEntryID) continue;
OUString sSourceURL; // Prefer temp file. It contains the changes against the original document! if (!info.OldTempURL.isEmpty())
sSourceURL = info.OldTempURL; elseif (!info.NewTempURL.isEmpty())
sSourceURL = info.NewTempURL; elseif (!info.OrgURL.isEmpty())
sSourceURL = info.OrgURL; else continue; // nothing real to save! An unmodified but new created document.
// TODO: Check eResult and react for errors (InteractionHandler!?) // Currently we ignore it ... // DON'T UPDATE THE CACHE OR REMOVE ANY TEMP. FILES FROM DISK. // That has to be forced from outside explicitly. // See implts_cleanUpWorkingEntry() for further details.
}
}
AutoRecovery::TDocumentList::iterator pIt = std::find_if(m_lDocCache.begin(), m_lDocCache.end(),
[&aParams](const AutoRecovery::TDocumentInfo& rInfo) { return rInfo.ID == aParams.m_nWorkingEntryID; }); if (pIt != m_lDocCache.end())
{
AutoRecovery::TDocumentInfo& rInfo = *pIt;
implts_flushConfigItem(rInfo, true); // sal_True => remove it from xml config!
m_lDocCache.erase(pIt);
}
}
AutoRecovery::EFailureSafeResult AutoRecovery::implts_copyFile(const OUString& sSource , const OUString& sTargetPath, const OUString& sTargetName)
{ // create content for the parent folder and call transfer on that content with the source content // and the destination file name as parameters
// exists session data ... => then we can't say, that these // data are valid for recovery. So we have to return sal_False then! if (bSessionData)
bRecoveryData = false;
aValue <<= bRecoveryData;
} break;
case AUTORECOVERY_PROPHANDLE_CRASHED :
aValue <<= officecfg::Office::Recovery::RecoveryInfo::Crashed::get(); break;
case AUTORECOVERY_PROPHANDLE_EXISTS_SESSIONDATA :
aValue <<= officecfg::Office::Recovery::RecoveryInfo::SessionData::get(); break;
}
}
sal_Int32 i = 0;
sal_Int32 c = xContainer->getCount();
for (i=0; i<c; ++i)
{
css::uno::Reference< css::frame::XFrame > xFrame; try
{
xContainer->getByIndex(i) >>= xFrame; if (!xFrame.is()) continue;
} // can happen in multithreaded environments, that frames was removed from the container during this loop runs! // Ignore it. catch(const css::lang::IndexOutOfBoundsException&)
{ continue;
}
// We are interested on visible documents only. // Note: It's n optional interface .-(
css::uno::Reference< css::awt::XWindow2 > xVisibleCheck(
xFrame->getContainerWindow(),
css::uno::UNO_QUERY); if (
(!xVisibleCheck.is() ) ||
(!xVisibleCheck->isVisible())
)
{ continue;
}
// extract the model from the frame. // Ignore "view only" frames, which does not have a model.
css::uno::Reference< css::frame::XController > xController;
css::uno::Reference< css::frame::XModel3 > xModel;
xController = xFrame->getController(); if (xController.is())
xModel.set( xController->getModel(), UNO_QUERY_THROW ); if (!xModel.is()) continue;
// insert model into cache ... // If the model is already well known inside cache // it's information set will be updated by asking the // model again for its new states.
implts_registerDocument(xModel);
}
} catch(const css::uno::RuntimeException&)
{ throw;
} catch(const css::uno::Exception&)
{
}
bool AutoRecovery::impl_enoughDiscSpace(sal_Int32 nRequiredSpace)
{ #ifdef SIMULATE_FULL_DISC return sal_False; #else// SIMULATE_FULL_DISC // In case an error occurs and we are not able to retrieve the needed information // it's better to "disable" the feature ShowErrorOnFullDisc ! // Otherwise we start a confusing process of error handling ...
// static void AutoRecovery::impl_establishProgress(const AutoRecovery::TDocumentInfo& rInfo ,
utl::MediaDescriptor& rArgs , const css::uno::Reference< css::frame::XFrame >& xNewFrame)
{ // external well known frame must be preferred (because it was created by ourself // for loading documents into this frame)! // But if no frame exists... we can try to locate it using any frame bound to the provided // document. Of course we must live without any frame in case the document does not exists at this // point. But this state should not occur. In such case xNewFrame should be valid ... hopefully .-)
css::uno::Reference< css::frame::XFrame > xFrame = xNewFrame; if (
(!xFrame.is() ) &&
(rInfo.Document.is())
)
{
css::uno::Reference< css::frame::XController > xController = rInfo.Document->getCurrentController(); if (xController.is())
xFrame = xController->getFrame();
}
// Any outside progress must be used ... // Only if there is no progress, we can create our own one.
css::uno::Reference< css::task::XStatusIndicator > xInternalProgress;
css::uno::Reference< css::task::XStatusIndicator > xExternalProgress = rArgs.getUnpackedValueOrDefault(
utl::MediaDescriptor::PROP_STATUSINDICATOR,
css::uno::Reference< css::task::XStatusIndicator >() );
// Normally a progress is set from outside (e.g. by the CrashSave/Recovery dialog, which uses our dispatch API). // But for a normal auto save we don't have such "external progress"... because this function is triggered by our own timer then. // In such case we must create our own progress ! if (
(! xExternalProgress.is()) &&
(xFrame.is() )
)
{
css::uno::Reference< css::task::XStatusIndicatorFactory > xProgressFactory(xFrame, css::uno::UNO_QUERY); if (xProgressFactory.is())
xInternalProgress = xProgressFactory->createStatusIndicator();
}
// HACK // An external provided progress (most given by the CrashSave/Recovery dialog) // must be preferred. But we know that some application filters query its own progress instance // at the frame method Frame::createStatusIndicator(). // So we use a two step mechanism: // 1) we set the progress inside the MediaDescriptor, which will be provided to the filter // 2) and we set a special Frame property, which overwrites the normal behaviour of Frame::createStatusIndicator .-) // But we suppress 2) in case we uses an internal progress. Because then it doesn't matter // if our applications make it wrong. In such case the internal progress resists at the same frame // and there is no need to forward progress activities to e.g. an outside dialog .-) if (
(xExternalProgress.is()) &&
(xFrame.is() )
)
{
css::uno::Reference< css::beans::XPropertySet > xFrameProps(xFrame, css::uno::UNO_QUERY); if (xFrameProps.is())
xFrameProps->setPropertyValue(FRAME_PROPNAME_ASCII_INDICATORINTERCEPTION, css::uno::Any(xExternalProgress));
}
// But inside the MediaDescriptor we must set our own create progress ... // in case there is not already another progress set.
rArgs.createItemIfMissing(utl::MediaDescriptor::PROP_STATUSINDICATOR, xInternalProgress);
}
// static void AutoRecovery::impl_forgetProgress(const AutoRecovery::TDocumentInfo& rInfo ,
utl::MediaDescriptor& rArgs , const css::uno::Reference< css::frame::XFrame >& xNewFrame)
{ // external well known frame must be preferred (because it was created by ourself // for loading documents into this frame)! // But if no frame exists... we can try to locate it using any frame bound to the provided // document. Of course we must live without any frame in case the document does not exists at this // point. But this state should not occur. In such case xNewFrame should be valid ... hopefully .-)
css::uno::Reference< css::frame::XFrame > xFrame = xNewFrame; if (
(!xFrame.is() ) &&
(rInfo.Document.is())
)
{
css::uno::Reference< css::frame::XController > xController = rInfo.Document->getCurrentController(); if (xController.is())
xFrame = xController->getFrame();
}
// stop progress interception on corresponding frame.
css::uno::Reference< css::beans::XPropertySet > xFrameProps(xFrame, css::uno::UNO_QUERY); if (xFrameProps.is())
xFrameProps->setPropertyValue(FRAME_PROPNAME_ASCII_INDICATORINTERCEPTION, css::uno::Any(css::uno::Reference< css::task::XStatusIndicator >()));
// forget progress inside list of arguments.
utl::MediaDescriptor::iterator pArg = rArgs.find(utl::MediaDescriptor::PROP_STATUSINDICATOR); if (pArg != rArgs.end())
{
rArgs.erase(pArg);
pArg = rArgs.end();
}
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.