It appears that the hsTArray memory management really sucks for smart
pointers like plKey. The crash mentioned at
http://forum.guildofwriters.org/viewtopic.php?f=117&t=6291 went away
immediately after switching plKeyCollector to an std::set.
This fixes various date formatting problems when building on a system where time_t is 64-bit (e.g. Visual Studio 2010), and, as a bonus, extends the range past 2038 on such systems.
The wire protocol is left at 32-bit for now, we might change that to 64 when other reasons to break compatibility have accumulated.
This should keep the avatar from getting stuck on some stupid stuff such
as on the pellet machine. A real fix would be to tweak the data, but we
cannot do that.
Author: Adam Johnson <AdamJohnso@gmail.com>
Date: Sun Mar 25 21:08:57 2018 -0400
Add workaround for mouse handling bug in Win10 FCU
Windows 10 Fall Creators Update changed the behavior of SetCursorPos,
which no longer sends the expected WM_MOUSEMOVE message when
recentering the cursor. This adds a function to detect affected
versions of Windows and flags the necessity to manually send our own
message.
This workaround currently only detects the known affected version of
Win10, in the hope that it will be fixed and no further attention
will be required. If this regression is not fixed, additional Win10
versions can be added.
Author: Adam Johnson <AdamJohnso@gmail.com>
Date: Tue Feb 5 18:59:20 2013 -0500
Reintroduce a fun hack
Yeah, looks like PhysX 2.6.4 has a bug with spawning stuff inside of
regions. So, let's bring back the hack that spawns the avatar at -2000
feet. The link-in process will set the correct position on the controller
and fire the appropriate detectors.
Author: Adam Johnson <AdamJohnso@gmail.com>
Date: Fri Feb 8 00:15:56 2013 -0500
Fix crazy camera stack issues
Turns out, it was an artifact of us suspending the simulation during links
and partly because of Cyan's late adding of the avatar controller to the
sim. Now, we add the avatar as soon as the age data is loaded. This causes
the camera stack to be populated with whatever garbage PhysX decides on,
then xJourneyClothsGen2 is free to set the real stack after we get all the
SDL from the server.
Verified to fix Teledahn oddness and not display a regression in Kemo.
Author: Adam Johnson <AdamJohnso@gmail.com>
Date: Mon Feb 6 21:17:08 2012 -0500
Fix an obvious memory leak.
This thing was pissing me off, which made me piss off a lot of other people,
which pissed off even more people, so the entire planet got pissed off and
nuclear warfare began. So, let's not do stupid crap and piss me off, okay?
Author: Adam Johnson
Date: Fri Feb 10 12:45:14 2012 -0500
More panic link fixes
Now, we check the armature mod to see if we're REALLY linked in (the screen
is not black). That way, we can be sure all delayed transforms and false
hits (like how we initially spawn into the AhnySphere04 panic region) don't
panic link us.
PhysX requires all of its actors to have cooked mesh data. This cooked
data is unfortunately not compatible with other versions of PhysX outside
of the minor branch. The format is also largely unknown. Therefore, it is
relevant to allow CWE prps to include an engine agnostic physical format
to reduce our dependencies on proprietary libraries.
All PhysX blobs thankfully begin with the magic string NXS\x01, so we now
abuse that by adding our own format that begins with the magic string
HSP\x01. These blobs are cooked on the fly while already cooked blobs are
passed directly on to PhysX.
Original commit: 392c5f5cad
author = "Adam Johnson <AdamJohnso@gmail.com>
and reenable as much USE_BINK_SDK stuff as empirically needed to get the Yeesha intro movie cave to disappear on its own, as before merging Cyan's Build 918.
Not the cleanest way, but I'm trying to minimize effort spent on this until we can get the proper solution, VP9 video support, from https://github.com/H-uru/Plasma/pull/458 .
In this case it now also works for non-owners of the age. Previously it only worked for owners because the ageInfoStruct does not contain the vault node ID, so it needed to be looked up somewhere, and that was in the AgesIOwnFolder.
Also, the text of the faded-out player list is re-rendered with alpha == 0 every time the mouse enters its area, do a little less needless work in that case.
The fix (adding adjustment of fRenderInfo.fDestPtr) made kRenderJustXForceLeft behave as intended (left-aligning the edge of the bitmap rather than the side bearing of the first character) but cuts off shadows in some places (e.g. at the left edges of "BUDDIES" and "NEIGHBORS" in the mini KI). To ensure enough space for the shadow, and considering that all content was developed and visually optimized with the bug in place, it seems better to preserve the buggy behavior and make kRenderJustXForceLeft work exactly like kRenderJustXLeft.
MarkD: My thought was for the long term was to put most of the differences in the project and solution configurations, so there would be less differences in the sources. EAX is a good example - I put the #define for EAX_SDK_AVAILABLE in the projects that used it with the thinking that the fans would create new projects that would not have EAX_SDK_AVAILABLE and would create and use a different configuration in the AllClient solution. Cyan would use the "Release" config and the fans use something like "Release-ou" to build with. But maybe there are just too many differences.
MarkD: We felt that the fan de-BINKing did not go far enough.
CW: I suspect (haven't tested yet) that it's the cause of the bug that we see on MOULa but did not on Minkata where the rotating wall behind the missing Yeesha intro video stays indefinitely, with no obvious way out for the player, until it occurs to them to use the ESC key
MarkD: Ah. I think you are right. An oversight that can be easily corrected - maybe in the python code so that Ian can see the rotating wall for a few seconds and then auto continue on.
In particular, the intro movie now exits immediately again rather than staying indefinitely.
The important difference is to send the completion callback in plBinkPlayer::NextFrame(), i.e. act as if we had reached the end of the movie.
Storing the filename is to keep plClient::IHandleMovieMsg() from deleting and recreating the plBinkPlayer on every message.
The changed return values are just to better match the previous behavior and probably don?t matter.
To better handle erroneous trigger events under the new timing method, use the number
of times the simulation has actually advanced instead of the number of evals received.
Cleaned up a bit.
Provides more accurate interaction with detectors when the avatar is moving up or down.
Also decreases the severity of the PhysX trigger hull bug in some detectors.
This change doesn't affect world collisions.
Adam Johnson
rarified
Christian Walther
cjkelly1
CyanWorlds
Skoader