For some reason there wasn't an entry in indra/llcommon/CMakeLists.txt to run
the tests in indra/llcommon/tests/lleventfilter_test.cpp. It seems likely that
at some point it existed, since all previous tests built and ran successfully.
In any case, (re-)add lleventfilter_test.cpp to the set of llcommon tests.
Also alphabetize them to make it easier to find a particular test invocation.
Also add new tests for LLEventThrottle.
To support this, refactor the concrete LLEventThrottle class into
LLEventThrottleBase containing all the tricky logic, with pure virtual
methods for access to LLTimer and LLEventTimeout, and an LLEventThrottle
subclass containing the LLTimer and LLEventTimeout instances and corresponding
implementations of the new pure virtual methods.
That permits us to introduce TestEventThrottle, an alternate subclass with
dummy implementations of the methods related to LLTimer and LLEventTimeout. In
particular, we can explicitly advance simulated realtime to simulate
particular LLTimer and LLEventTimeout behaviors.
Finally, introduce Concat, a test LLEventPump listener class whose function is
to concatenate received string event data into a composite string so we can
readily test for particular sequences of events.
Drake points out that the OS X 64-bit-capable memory-query APIs recommended in
comments by some long-ago maintainer are by now themselves obsolete. He
offered this patch to update us to current macOS memory APIs.
For the time being we're still compiling for production with C++03. Although
assigning an initializer list to a vector is valid C++11, in C++03 mode clang
rejects it.
LLInstanceTracker<T> performs validation in ~LLInstanceTracker(). Normally
validation failure logs an error and terminates the program, which is fine. In
the test executable, though, we want validation failure to throw an exception
instead so we can catch it and continue testing other failure conditions. But
since destructors in C++11 are implicitly noexcept(true), that exception never
made it out of ~LLInstanceTracker(): it crashed the test program instead.
Declaring ~LLInstanceTracker() noexcept(false) solves that, allowing the test
program to catch the exception and continue.
However, if we unconditionally declare that, then every destructor anywhere in
the inheritance hierarchy for any LLInstanceTracker subclass must also be
noexcept(false)! That's way too pervasive, especially for functionality we
only need (or want) in a specific test executable.
Instead, make the CMake macros LL_ADD_PROJECT_UNIT_TESTS() and
LL_ADD_INTEGRATION_TEST() -- with which we define all viewer build-time tests
-- define two new command-line macros: LL_TEST=testname and LL_TEST_testname.
That way, preprocessor logic in a header file can detect whether it's being
compiled for production code or for a test executable.
(While at it, encapsulate in a new GET_OPT_SOURCE_FILE_PROPERTY() CMake macro
an ugly repetitive pattern. The builtin GET_SOURCE_FILE_PROPERTY() sets the
target variable to "NOTFOUND" -- rather than an empty string -- if the
specified property wasn't set. Every call to GET_SOURCE_FILE_PROPERTY() in
LL_ADD_PROJECT_UNIT_TESTS() was followed by a test for NOTFOUND and an
assignment to "". Wrap all that in a macro whose 'unset' value is "".)
Now llinstancetracker.h can detect when we're building the LLInstanceTracker
unit test executable, and *only then* declare ~LLInstanceTracker() as
noexcept(false). We #define LLINSTANCETRACKER_DTOR_NOEXCEPT to expand either
empty or noexcept(false), also detecting clang in C++11 mode. (It all works
fine without noexcept(false) until we turn on C++11 mode.)
We also use that macro for the StatBase class in lltrace.h. Turns out some of
the infrastructure headers required for tests in general, including the
LLInstanceTracker test, use LLInstanceTracker. Fortunately that appears to be
the only other class we must annotate this way for the LLInstanceTracker tests.
The LLMemory method getCurrentRSS() is defined to return the "resident set
size," but in fact on Windows it returns the WorkingSetSize -- and that's
actually what callers want from it: the total memory consumed by the
application for statistics purposes. It's not really clear what users gain by
knowing how much of that is resident in real memory, versus the total
consumption. So despite the commentation and the method name itself, on Mac
make it return the virtual size consumed.
Whatever we were trying to do with LLSharedLibs.cmake hasn't worked on the Mac
for a long time, and trying to fix it only digs up more problems. Skip it:
we've already worked around it.
Update the media_plugins_example CMakeLists.txt to eliminate some CMake
non-existent dependency warnings.
Evidently the Mac implementation of LLMemory::getCurrentRSS() goes back to
OS X 10.3, because there was a helpful comment of the form:
------
The API used here is not capable of dealing with 64-bit memory sizes, but is
available before 10.4.
Once we start requiring 10.4, we can use the updated API, which looks like
this:
[new current implementation]
Of course, this doesn't gain us anything unless we start building the viewer
as a 64-bit executable, since that's the only way for our memory allocation to
exceed 2^32.
------
Hey, guess what, we're building 64-bit viewers now!
Thank you, whoever thoughtfully noted that, both for calling out the issue and
sparing us the research. (The comment goes back to Subversion days, so hg
blame shows only the merge-to-release changeset.)
There were two distinct LLMemory methods getCurrentRSS() and
getWorkingSetSize(). It was pointless to have both: on Windows they were
completely redundant; on other platforms getWorkingSetSize() always returned
0. (Amusingly, though the Windows implementations both made exactly the same
GetProcessMemoryInfo() call and used exactly the same logic, the code was
different in the two -- as though the second was implemented without awareness
of the first, even though they were adjacent in the source file.)
One of the actual MAINT-6996 problems was due to the fact that
getWorkingSetSize() returned U32, where getCurrentRSS() returns U64. In other
words, getWorkingSetSize() was both useless *and* wrong. Remove it, and change
its one call to getCurrentRSS() instead.
The other culprit was that in several places, the 64-bit WorkingSetSize
returned by the Windows GetProcessMemoryInfo() call (and by getCurrentRSS())
was explicitly cast to a 32-bit data type. That works only when explicitly or
implicitly (using LLUnits type conversion) scaling the value to kilobytes or
megabytes. When the size in bytes is desired, use 64-bit types instead.
In addition to the symptoms, LLMemory was overdue for a bit of cleanup.
There was a 16K block of memory called reserveMem, the comment on which read:
"reserve 16K for out of memory error handling." Yet *nothing* was ever done
with that block! If it were going to be useful, one would think someone would
at some point explicitly free the block. In fact there was a method
freeReserve(), apparently for just that purpose -- which was never called. As
things stood, reserveMem served only to *prevent* the viewer from ever using
that chunk of memory. Remove reserveMem and the unused freeReserve().
The only function of initClass() and cleanupClass() was to allocate and free
reserveMem. Remove initClass(), cleanupClass() and the LLCommon calls to them.
In a similar vein, there was an LLMemoryInfo::getPhysicalMemoryClamped()
method that returned U32Bytes. Its job was simply to return a size in bytes
that could fit into a U32 data type, returning U32_MAX if the 64-bit value
exceeded 4GB. Eliminate that; change all its calls to getPhysicalMemoryKB()
(which getPhysicalMemoryClamped() used internally anyway). We no longer care
about any platform that cannot handle 64-bit data types.
Glenn Glazer