| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
build.sh LogScan greps for "error:" (among other things) so removing the colon
from the test name "syntax_error" should help.
|
|
|
|
|
|
Introduce LLCoros::setStackSize(), with a compile-time default value we hope
we never have to use. Make LLAppViewer call it with the value of the new
settings variable CoroutineStackSize as soon as we've read settings files.
(While we're at it, notify interested parties that we've read settings files.)
Give CoroutineStackSize a default value four times the previous default stack
size. Make LLCoros::launch() pass the saved stack size to each new coroutine
instance.
Re-enable lleventcoro integration test. Use LLSDMap() construct rather than
LLSD::insert(), which used to return the modified object but is now void.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Normalize deadman timer's args on U64/F64. Internals remain the
same. Modify mesh to collect and output enhanced CPU metrics.
|
|
Integrated as a ctor-time option to LLDeadmanTimer and have mesh
use this mode for the stats I'm gathering.
|
|
Windows resolution isn't good enough for a strictly increasing
time test in the unit tests.
|
|
The http_texture_load example program has some cpu usage gathering
tools that should be generally useful and specifically for the
deadman switch. Port these into llcommon into new all-static
class LLProcInfo. Add unit test, etc.
|
|
Timer interface violated my design rules and I paid for it
with clumsiness and silent errors. Cleaned it up mainly
removing the evil default values. Found better integration
points in the mesh downloader and it's producing fairly
consistent numbers on the MeshTest2 test region (about
5500 downloads, ~90 seconds, +/- 10 seconds). Will review
with davep and do an early timer stop on teleport which
invalidates a timing sequence.
|
|
doesn't allow negative offsetting for unit tests. Had to keep
things positive. Could do windows-specific test cases but I'm
hoping to dump lltimer from the implementation later.
|
|
the deadman's timer. Found some bugs, dig some cleanup and
documented a few things. Definitely want to get rid of the
U64/F64 interfaces at sometime but this is a good start.
|
|
Will be used for mesh, inventory, etc., operation markers.
|
|
|
|
|
|
|
|
from this tree
|
|
In autobuild.xml, specify today's build of the Boost package that includes the
Boost.Context library, and whose boost::dcoroutines library uses Boost.Context
exclusively instead of its previous context-switching underpinnings (source of
the ucontext.h dependency).
Add BOOST_CONTEXT_LIBRARY to Boost.cmake and Copy3rdPartyLibs.cmake. Link it
with the viewer and with the lllogin.cpp test executable.
Track new Boost package convention that our (early, unofficial) Boost.Coroutine
library is now accessed as boost/dcoroutine/etc.h and boost::dcoroutines::etc.
Remove #include <boost/coroutine/coroutine.hpp> from
llviewerprecompiledheaders.h and lllogin.cpp: old rule that Boost.Coroutine
header must be #included before anything else that might use ucontext.h is
gone now that we no longer depend on ucontext.h. In fact remove
-D_XOPEN_SOURCE in 00-Common.cmake because that was inserted specifically to
work around a known problem with the ucontext.h facilities.
|
|
|
|
* Removed no longer used unpack_bufsize from bitpack_test.cpp
* Added llviewertexture_stub.cpp to the newview tests directory to fix llworldmap_test.cpp and llworldmipmap_test.cpp linker errors.
|
|
alignment issue in llAppearance.
|
|
|
|
|
|
|
|
The LLURI::buildHTTP() overloads that take an LLSD 'path' accept 'undefined',
LLSD::String and (LLSD::Array of LLSD::String). A sequence of path components
passed in an Array is constructed into a slash-separated path. There are unit
tests in lluri_test.cpp to exercise that case.
To my amazement, there were NO unit tests covering the case of an LLSD::String
path. The code for that case escaped and appended the entire passed string.
While that might be fine for a 'path' consisting of a single undecorated path
component, the available documentation does not forbid one from passing a path
containing slashes as well. But this had the dubious effect of replacing every
slash with %2F.
In particular, decomposing a URL string with one LLURI instance and
constructing another like it using LLURI::buildHTTP() was not symmetrical.
Having consulted with Richard, I made the string-path logic a bit more nuanced:
- The passed path string is split on slashes. Every path component is
individually escaped, then recombined with slashes into the final path.
- Duplicate slashes are eliminated.
- The presence or absence of a trailing slash in the original path string is
carefully respected.
Now that we've nailed down how it ought to behave -- added unit tests to
ensure that it DOES behave that way!!
|
|
Big delta was converting the new texture debugger support code
to the new library. Viewer manifest should probably get an eyeball
before release.
|
|
|
|
That is, when the underlying LLError::Settings object is destroyed -- possibly
at termination, possibly on LLError::restoreSettings() -- the passed Recorder*
is deleted.
There was much existing code that seemed as unaware of this alarming fact as I
was myself. Passing to addRecorder() a pointer to a stack object, or to a
member of some other object, is just Bad. It might be preferable to make
addRecorder() accept std::auto_ptr<Recorder> to make the ownership transfer
more explicit -- or even boost::shared_ptr<Recorder> instead, which would
allow the caller to either forget or retain the passed Recorder.
This preliminary pass retains the Recorder* dumb pointer API, but documents
the ownership issue, and eliminates known instances of passing pointers to
anything but a standalone heap Recorder subclass object.
|
|
This test must not be subject to spurious environmental failures, else some
kind soul will disable it entirely. We observe that APR specifies a hard-coded
buffer size of 64Kbytes for pipe creation -- use that and cross fingers.
|
|
Sigh, the rejoicing was premature.
|
|
If in fact we've managed to fix the APR bug writing to a Windows named pipe,
it should no longer be necessary to try to work around it by testing with a
much smaller data volume on Windows!
|
|
Ideally we'd love to be able to nail the underlying bug, but log output
suggests it may actually go all the way down to the OS level. To move forward,
try to bypass it.
|
|
We want to write a robust test that consistently works. On Windows, that
appears to require constraining the max message size. I, the coder, could try
submitting test runs of varying sizes to TC until I found a size that works...
but that could take quite a while. If I were clever, I might even use a manual
binary search. But computers are good at binary searching; there are even
prepackaged algorithms in the STL. If I were cleverer still, I could make the
test program itself search for size that works.
|
|
Apparently, at least on Mac, there are circumstances in which the very-large-
message test can take several times longer than normal, yet still complete
successfully. This is always the problem with timeouts: does timeout
expiration mean that the code in question is actually hung, or would it
complete if given a bit longer?
If very-large-message test fails, retry a few times with smaller sizes to try
to find a size at which the test runs reliably. The default size, ca 1MB, is
intended to be substantially larger than anything we'll encounter in the wild.
Is that "unreasonably" large? Is there a "reasonable" size at which the test
could consistently pass? Is that "reasonable" size still larger than what we
expect to encounter in practice? Need more information, hence this code.
|
|
Otherwise, a stuck child process could potentially hang the test, and thus the
whole viewer build.
|
|
It seems that under certain circumstances, write logic was duplicating a chunk
of the data being streamed down our pipe. But as this condition is only driven
with a very large data stream, eyeballing that data stream is tedious. Add
code to compare the raw received data with the expected stream, reporting
where and how they first differ.
|
|
That lets us reliably declare the operator<<() free function inline, which
permits multiple translation units in the same executable to #include
"wrapllerrs.h".
|
|
While we're accumulating the 'length:' prefix, the present socket-based logic
reads 20 characters, then reads 'length' more, then discards any excess (in
case the whole 'length:data' packet ends up being less than 20 characters).
That's probably a bug: whatever characters follow that packet, however short
it may be, are probably the 'length:' prefix of the next packet. We probably
only get away with it because we probably never send packets that short.
Earlier llleap_test.cpp plugin logic still read 20 characters, then, if there
were any left after the present packet, cached them as the start of the next
packet. This is probably more correct, but complicated. Easier just to read
individual characters until we've seen 'length:', then try for exactly the
specified length over however many reads that requires.
|
|
In load testing, we have observed intermittent failures on Windows in which
LLSDNotationStreamer into std::ostringstream seems to bump into a hard limit
of 1048590 bytes. ostringstream reports that much buffered data and returns
that much -- even though, on examination, the notation-serialized stream is
incomplete at that point. It's our intention to load-test LLLeap and
LLProcess, not the local iostream implementation; we hope that this kind of
data volume is comfortably greater than actual usage. Back off the
load-testing max size a bit.
|
|
New llleap_test.cpp load testing turned up Windows issue in which plugin
process received corrupt packet, producing LLSDParseError. Add code to dump
the bad packet in that case -- but if LLSDParseError is willing to state the
offset of the problem, not ALL of the packet.
Quiet MSVC warning about little internal base class needing virtual destructor.
|
|
|
