The crash was caused by erroneous getting of month name from vector with week day names in LLStringUtil::formatDatetime().
This code woth introduced in June, so though it didn't work properly, it didn't cause the crash(cause June is 5th month). But when
number of current month exceeded number of days in week(this happened in August cause it is 8th) code started getting 8th element from
vector with 7. This caused the crash. It reproduced only on Japanese locale because only there code that caused it was used(see STORM-177
for details). This changeset seems to fix STORM-177 too.
- Used vector with months names where it should be.
Changes:
- Added support for formatting day of the month without leading zero ("sday").
- Changed date format in place profile (landmark info) and in the top status bar
according to bug reporter's request.
Technical details:
Actually implementation of strftime() in Linux and Windows supports stripping the
leading zero (with "%-d" and "%#d" respectively).
But that's not supported in MacOSX, so I had to reimplement it.
Reviewed by Sergey Litovchuk at https://codereview.productengine.com/secondlife/r/842/
--HG--
branch : product-engine
Assumprion is: wide char buffer requires not more than input string length plus one for a null terminator.
Reviewed by Richard Nelson at https://codereview.productengine.com/secondlife/r/775/
--HG--
branch : product-engine
Problem on Windows:
==================
LLPanelMainInventory::updateItemcountText() formats number using viewer locale.
non-break space is detected as unknown symbols while converting utf8str_to_wstring when formatted text is set to LLTextBox.
FIX:
===
Added converting of string to multi-byte string and then to utf8 string while formatting on Windows.
created opposite to "ll_convert_wide_to_string" function "ll_convert_string_to_wide" and helper function to call both of them.
It is used now to convert result of formatted string while formatting integer number in locale.
Fix affects Windows only.
Reviewed by Richard Nelson at https://codereview.productengine.com/secondlife/r/775/
--HG--
branch : product-engine
Calling std::locale("fr_FR.UTF-8") crashes on Linux and Mac. Or
rather, it throws an exception when it doesn't know the locale and we
didn't handle the exception. I now catch the exception and output an
error rather than crash.
Note, this happened because of change 703f3bcf7069, which made us
actually pass a real locale string instead of just "C". So, we were
never actually supporting a locale for LLStringUtil::formatNumber().
There is therefore an open task of making formatNumber() actually
respect the locale. I'll report a separate JIRA to capture that task.
Problem:
* English locale was set for all languages.
* Specifying a correct locale didn't affect anything, including date/time formatting.
My investigation has shown that LLStringUtil was instantiated twice: in the
main binary and in libllcommon.so.
Because LLStringUtil::setLocale() was called from newview and getLocale()
was called from llcommon, they effectively used *different* instances of
LLStringUtil::sLocale. Hence getLocale() always returned empty string.
This seems to be caused by get/setLocale() methods not being dllexported.
The fix instantiates get/setLocale() and sLocale in llcommon and exposes
them to use from newview (i.e. prevents multiple instantiation).
Besides, I specified correct locale names for all languages and platforms.
Reviewed by Leyla: https://codereview.productengine.com/secondlife/r/104/
--HG--
branch : product-engine
* Bugs: EXT-888 EXT-866 EXT-861 EXT-858 EXT-864 EXT-875 EXT-884 EXT-718 EXT-786 EXT-885 EXT-910 EXT-845 EXT-312 EXT-823 EXT-868
* New Development: EXT-748 EXT-863 EXT-835
QA: Please test Recent List to verify it has no troubles.
svn merge -r107352:108210
svn+ssh://svn.lindenlab.com/svn/linden/branches/server/server-1.25
Merge conflicts were resolved:
scripts/start-server : one conflict, kept the "working" version
indra/llcommon/llversionserver.h : reverted
indra/newview/Info-SecondLife.plist : reverted
indra/newview/res/viewerRes.rc : reverted
indra/newsim/llagentinfo.cpp : kept *both*... "working" had
is_agent_online, which nothing else had, and the merge had
isGodAllowed(), part of SEC resolution
indra/backbone/agent_router.py : kartic & rdw merged and reviewed
using brains and things
indra/newsim/llsimhttpresponder.cpp : doesn't exist in trunk, and
that's OK (even though there was a change for it)
Effectively:
svn merge -r 87048:88773 svn+ssh://svn.lindenlab.com/svn/linden/branches/Branch_1-22-Server
* Plus some gcc4.x fixes
In detail:
* Bump version numbers to 1.22.0 Server
* Updated files for enabled conductor.
* Added tarball pulling during deploy for central servers.
* DEV-15160 Logins not remembering start location on beta grid
* DEV-15167 Random failure with set-classified-stats
* DEV-15193 Viewer Passing Through Unsanitized Strings to DB when creating classifieds.
* Add --location to the curl invocation in fetch_agni_ws_snapshot
* QAR-609 Some Havok4 bug fixes
* QAR-460 Automatic XUI selection & Kick-ass Linux Fonts MergeMe
* DEV-14704 1.21.1.86182 revealing hidden groups in profile
* DEV-14641 Change automated Subject Line messages for IM-to-Email to avoid SPAM association
* DEV-15387 Write unit test for LLStringFn::strip_invalid_xml().
* Changed getpass.getuser() to os.getuid() as getuser() was returning the saved-UID of the process (root user, which kic
ks off parabuild as parabuild u
ser). We want os.getuid() as it correctly returns the parabuild user's uid. / Stupid mistake using os.getuid().. .fixed
by using the pwd module to return the unix username given the userid from os.getuid()
* QAR-623 More single ref attachment awesomeness testing
* Committed temporary fix for running dwell and query-cache on deneb.lindenlab.com, since it's in the DFW colo with the
current asset cluster.
* DEV-15783 Groups not shown in search wont appear in your profile while using webservice
* QAR-642 Even more single ref attachment awesomeness testing
From Branch_1-21-Server (which didn't go directly into release)
* addinng the phputf8 library found from http://phputf8.sourceforge.net/ and liscensed LGPL. For DEV-15202
* Checking in a fix for settle balance portion proc - the event query was moved.
* fix for DEV-15202 - replace non-utf8 with REPLACEMENT CHAR. See jira for more info.
* Checking return value from /proc/vault/charge, must get back a string.
* removed code used for the browser hud AB test.
* Changed TTL from stupid default of 1234 to 5
* DEV-15788 Fixing crashloganalyzer.py cleanup of unparseable crash reports.
* DEV-15737 offline scripted IMs not working inprocess_im_using_ws
Merge conflict resolutions/fixes include:
* Addition of indra/test/llstring_tut.cpp to indra/test/test.vcproj migrated to CMakeLists.txt
* Changes to BuildParams, scripts/automated_build_scripts/* reviewed w/ cg
* Changes to deploy scripts (scripts/farm, install_indra.py, farm_deploy) based on the 1.22 changes except for mkpersist
dir addition (retained from release)
* Changes to scripts/crash_report/crashloganalyzer.py retained from 1.22, per Mani
* scripts/generate_indra_xml.py, etc/services-station.dtd, indra/newsim/llagentinfo.cpp analyzed carefully by josh
* indra/llcommon/llstring.h required fix for client build ("" --> std::basic_string<T>()) in LLStringBase<T> initializer
* indra/llphysics/abstract/llphysicsshape.h #include syntax changed from <llmath/llvolume.h> to "llvolume.h"
Psst, don't tell anyone, but the dataserver-is-deprecated
HAVOK4 IN TEH HOUSE!!11!!ONE!!
If it is broken blame Joel for not fixing the loginassetdatabaseinventorygroupIM server instead of working on this.
QAR-448
Richard Linden