The previous hack wasn't thread-safe: read-only access would
access the reference counter multiple times at the same time,
which therefore would have to be thread-local to ever work.
The current solution just disables the calls to lock/unlock
for copyconstructed objects, which works if the copyconstructed
object isn't used anymore after the original is destructed.
This is the case then the copy construction only happens
upon passing a temporary to a function, which is the case.
g++ 4.2 (and possibly earlier) apparently call a copy
constructor when passing a temporary to a function
that takes a const reference. Added code to allow
copy-constructing the AI*Access classes for this
compiler.
g++-4.2.x also bails out when it encounters files that
do not end on a newline. So, also added those where
they were missing.
Wrapped gSettings and the global objects returned by Settings::get() and
Globals::get() in AIThreadSafe, forcing thread-safe access.
This solves the problem of possible corruption of the various LevelMap's
in LLError::Settings due to thread unsafe accesses.
This involves making libllcommon shared, a change
that was not finished for windows/Mac.
More changes are needed to indra/newview/viewer_manifest.py
for those two operating systems, as well as to
indra/copy_win_scripts.
I did not test this on linux 32bit, nor non-standalone
on that OS, either.
I did not check voice (which will very likely not work
anyway on 64bit). And currently the webkit plugin doesn't
work for me, but I'll fix that in a different commit.
Most the changes are due to InventoryObjectList* changing to LLInventoryObject::object_list_t*, LLInventoryItem::II_FLAGS* changing to LLInventoryItemFlags::II_FLAGS* and also const'ing.
Certain perms for certain asset types(callcard&landmarks) have been laxed, as per LL's V2.
LLInventoryType now does lookups mostly though new lldictionary class.
LLLandmark using boost for callbacks, instead of custom class structure.
Excluded llareslistener, as that appears to only be present for unit-testing
Excluded new SSL methods because, well, they don't work right reliably in v2 for me
