Always search for python3[.exe] instead of plain 'python'. macOS Monterey no
longer bundles Python 2 at all.
Explicitly make PYTHON_EXECUTABLE a cached value so if the user edits it in
CMakeCache.txt, it won't be overwritten by indra/cmake/Python.cmake.
Do NOT set DYLD_LIBRARY_PATH for test executables! That has Bad Effects, as
discussed in https://stackoverflow.com/q/73418423/5533635. Instead, create
symlinks from build-mumble/sharedlibs/Resources -> Release/Resources and from
build-mumble/test/Resources -> ../sharedlibs/Release/Resources. For test
executables in sharedlibs/RelWithDebInfo and test/RelWithDebInfo, this
supports our dylibs' baked-in load path @executable_path/../Resources. That
load path assumes running in a standard app bundle (which the viewer in fact
does), but we've been avoiding creating an app bundle for every test program.
These symlinks allow us to continue doing that while avoiding
DYLD_LIBRARY_PATH.
Add indra/llcommon/apply.h. The LL::apply() function and its wrapper macro
VAPPLY were very useful in diagnosing the problem.
Tweak llleap_test.cpp. This source was modified extensively for diagnostic
purposes; these are the small improvements that remain.
This changeset makes it possible to build the Second Life viewer using
Python 3. It is designed to be used with an equivalent Autobuild branch
so that a developer can compile without needing Python 2 on their
machine.
Breaking change: Python 2 support ending
Rather than supporting two versions of Python, including one that was
discontinued at the beginning of the year, this branch focuses on
pouring future effort into Python 3 only. As a result, scripts do not
need to be backwards compatible. This means that build environments,
be they on personal computers and on build agents, need to have a
compatible interpreter.
Notes
- SLVersionChecker will still use Python 2 on macOS
- Fixed the message template url used by template_verifier.py
Turns out that one of our WorkQueue integration tests was relying on the
incorrect runFor() behavior that we just fixed, so the test broke. Now that
runFor() doesn't wait around for work to be posted, use an explicit wait loop
instead.
To support this, add LLCond::get(functor), where functor must accept a const
reference to the stored data. This new get() returns whatever the functor
returns, allowing a caller to peek at the stored data.
Also use universal references for all remaining LLCond functor arguments.
Reverting a merge is sticky: it tells git you never want to see that branch
again. Merging the DRTVWR-546 branch, which contained the revert, into the
glthread branch undid much of the development work on that branch. To restore
it we must revert the revert.
This reverts commit 029b41c041.
A typical WorkQueue has a string name, which can be used to find it to post
work to it. "Work" is a nullary callable.
WorkQueue is a multi-producer, multi-consumer thread-safe queue: multiple
threads can service the WorkQueue, multiple threads can post work to it.
Work can be scheduled in the future by submitting with a timestamp. In
addition, a given work item can be scheduled to run on a recurring basis.
A requesting thread servicing a WorkQueue of its own, such as the viewer's
main thread, can submit work to another WorkQueue along with a callback to be
passed the result (of arbitrary type) of the first work item. The callback is
posted to the originating WorkQueue, permitting safe data exchange between
participating threads.
Methods are provided for different kinds of servicing threads. runUntilClose()
is useful for a simple worker thread. runFor(duration) devotes no more than a
specified time slice to that WorkQueue, e.g. for use by the main thread.
ThreadSafeSchedule::tryPopUntil() (and therefore tryPopFor()) was simply
delegating to LLThreadSafeQueue::tryPopUntil(), with an adjusted timeout since
we want to wake up as soon as the head item, if any, becomes ready. But then
we have to loop back to retry the pop to actually deal with that head item.
In addition, ThreadSafeSchedule::popWithTime() was spinning rather than
properly blocking on a timed condition variable. Fixed.
ThreadSafeSchedule orders its items by timestamp, which can be passed either
implicitly or explicitly. The timestamp specifies earliest delivery time: an
item cannot be popped until that time.
Add initial tests.
Tweak the LLThreadSafeQueue base class to support ThreadSafeSchedule:
introduce virtual canPop() method to report whether the current head item is
available to pop. The base class unconditionally says yes, ThreadSafeSchedule
says it depends on whether its timestamp is still in the future.
This replaces the protected pop_() overload accepting a predicate. Rather than
explicitly passing a predicate through a couple levels of function call, use
canPop() at the level it matters. Runtime behavior that varies depending on
an object's leaf class is what virtual functions were invented for.
Give pop_() a three-state enum return so pop() can distinguish between "closed
and empty" (throws exception) versus "closed, not yet drained because we're
not yet ready to pop the head item" (waits).
Also break out protected tryPopUntil_() method, the body logic of
tryPopUntil(). The public method locks the data structure, the protected
method requires that its caller has already done so.
Add chrono.h with a more full-featured LL::time_point_cast() function than the
one found in <chrono>, which only converts between time_point durations, not
between time_points based on different clocks.
These functions allow prepending or removing an item at the left end of an
arbitrary tuple -- for instance, to add a sequence key to a caller's data,
then remove it again when delivering the original tuple.
Introduce Oz's LLERROR_CRASH macro analogous to the old LLError::crashAndLoop()
function. Change LL_ENDL macro so that, after calling flush(), if the CallSite
is for LEVEL_ERROR, we invoke LLERROR_CRASH right there.
Change the meaning of LLError::FatalFunction. It used to be responsible for
the actual crash (hence crashAndLoop()). Now, instead, its role is to disrupt
control flow in some other way if you DON'T want to crash: throw an exception,
or call exit() or some such. Any FatalFunction that returns normally will fall
into the new crash in LL_ENDL.
Accordingly, the new default FatalFunction is a no-op lambda. This eliminates
the need to test for empty (not set) FatalFunction in Log::flush().
Remove LLError::crashAndLoop() because the official LL_ERRS crash is now in
LL_ENDL.
One of the two common use cases for setFatalFunction() used to be to intercept
control in the last moments before crashing -- not to crash or to avoid
crashing, but to capture the LL_ERRS message in some way. Especially when
that's temporary, though (e.g. LLLeap), saving and restoring the previous
FatalFunction only works when the lifespans of the relevant objects are
strictly LIFO.
Either way, that's a misuse of FatalFunction. Fortunately the Recorder
mechanism exactly addresses that case. Introduce a GenericRecorder template
subclass, with LLError::addGenericRecorder(callable) that accepts a callable
with suitable (level, message) signature, instantiates a GenericRecorder, adds
it to the logging machinery and returns the RecorderPtr for possible later use
with removeRecorder().
Change llappviewer.cpp's errorCallback() to an addGenericRecorder() callable.
Its role was simply to update gDebugInfo["FatalMessage"] with the LL_ERRS
message, then call writeDebugInfo(), before calling crashAndLoop() to finish
crashing. Remove the crashAndLoop() call, retaining the gDebugInfo logic. Pass
errorCallback() to LLError::addGenericRecorder() instead of setFatalFunction().
Oddly, errorCallback()'s crashAndLoop() call was conditional on a compile-time
SHADER_CRASH_NONFATAL symbol. The new mechanism provides no way to support
SHADER_CRASH_NONFATAL -- it is a Bad Idea to return normally from any LL_ERRS
invocation!
Rename LLLeapImpl::fatalFunction() to onError(). Instead of passing it to
LLError::setFatalFunction(), pass it to addGenericRecorder(). Capture the
returned RecorderPtr in mRecorder, replacing mPrevFatalFunction. Then
~LLLeapImpl() calls removeRecorder(mRecorder) instead of restoring
mPrevFatalFunction (which, as noted above, was order-sensitive).
Of course, every enabled Recorder is called with every log message. onError()
and errorCallback() must specifically test for calls with LEVEL_ERROR.
LLSingletonBase::logerrs() used to call LLError::getFatalFunction(), check the
return and call it if non-empty, else call LLError::crashAndLoop(). Replace
all that with LLERROR_CRASH.
Remove from llappviewer.cpp the watchdog_llerrs_callback() and
watchdog_killer_callback() functions. watchdog_killer_callback(), passed to
Watchdog::init(), used to setFatalFunction(watchdog_llerrs_callback) and then
invoke LL_ERRS() -- which seems a bit roundabout. watchdog_llerrs_callback(),
in turn, replicated much of the logic in the primary errorCallback() function
before replicating the crash from llwatchdog.cpp's default_killer_callback().
Instead, pass LLWatchdog::init() a lambda that invokes the LL_ERRS() message
formerly found in watchdog_killer_callback(). It no longer needs to override
FatalFunction with watchdog_llerrs_callback() because errorCallback() will
still be called as a Recorder, obviating watchdog_llerrs_callback()'s first
half; and LL_ENDL will handle the crash, obviating the second half.
Remove from llappviewer.cpp the static fast_exit() function, which was simply
an alias for _exit() acceptable to boost::bind(). Use a lambda directly
calling _exit() instead of using boost::bind() at all.
In the CaptureLog class in llcommon/tests/wrapllerrs.h, instead of statically
referencing the wouldHaveCrashed() function from test.cpp, simply save and
restore the current FatalFunction across the LLError::saveAndResetSettings()
call.
llerror_test.cpp calls setFatalFunction(fatalCall), where fatalCall() was a
function that simply set a fatalWasCalled bool rather than actually crashing
in any way. Of course, that implementation would now lead to crashing the test
program. Make fatalCall() throw a new FatalWasCalled exception. Introduce a
CATCH(LL_ERRS("tag"), "message") macro that expands to:
LL_ERRS("tag") << "message" << LL_ENDL;
within a try/catch block that catches FatalWasCalled and sets the same bool.
Change all existing LL_ERRS() in llerror_test.cpp to corresponding CATCH()
calls. In fact there's also an LL_DEBUGS(bad tag) invocation that exercises an
LL_ERRS internal to llerror.cpp; wrap that too.
LLSDNotationFormatter (also LLSDNotationStreamer that uses it, plus
operator<<(std::ostream&, const LLSD&) that uses LLSDNotationStreamer) is most
useful for displaying LLSD to a human, e.g. for logging. Having the default
dump raw binary bytes into the log file is not only suboptimal, it can
truncate the output if one of those bytes is '\0'. (This is a problem with the
logging subsystem, but that's a story for another day.)
Use OPTIONS_PRETTY_BINARY wherever there is a default LLSDFormatter
::EFormatterOptions argument.
Also, allow setting LLSDFormatter subclass boolalpha(), realFormat() and
format(options) using optional constructor arguments. Naturally, each subclass
that supports this must accept and forward these constructor arguments to its
LLSDFormatter base class constructor.
Fix a couple bugs in LLSDNotationFormatter::format_impl() for an LLSD::Binary
value with OPTIONS_PRETTY_BINARY:
- The code unconditionally emitted a b(len) type prefix followed by either raw
binary or hex, depending on the option flag. OPTIONS_PRETTY_BINARY caused it
to emit "0x" before the hex representation of the data. This is wrong in
that it can't be read back by either the C++ or the Python LLSD parser.
Correct OPTIONS_PRETTY_BINARY formatting consists of b16"hex digits" rather
than b(len)"raw bytes".
- Although the code did set hex mode, it didn't set either the field width or
the fill character, so that a byte value less than 16 would emit a single
digit rather than two.
Instead of having one LLSDFormatter::format() method with an optional options
argument, declare two overloads. The format() overload without options passes
the mOptions data member to the overload accepting options.
Refactor the LLSDFormatter family, hoisting the recursive format_impl() method
(accepting level) to a pure virtual method at LLSDFormatter base-class level.
Most subclasses therefore need not override either base-class format() method,
only format_impl(). In fact the short format() overload isn't even virtual.
Consistently use LLSDFormatter::EFormatterOptions enum as the options
parameter wherever such options are accepted.
llmainthreadtask_test builds in a Sync timeout to keep build-time tests from
hanging. That timeout was set to 2000ms, which seems as though it ought to be
plenty enough time for a process with only 2 threads to exchange data between
them. But on TeamCity EC2 Windows build hosts, sometimes we hit that timeout
and fail. Extend it to try to improve the robustness of builds, even though
the possibility of a production viewer blocking for that long for anything
seems worrisome. (Fortunately the production viewer does not use Sync.)
Ansariel