On Week 20

Posted by skyjake in Blog on May 23, 2011

skyjake:

I was mostly focusing on the animation of remote players. I sorted out the handling of clients’ forward and side move values (which used to be part of the ticcmd struct) by including those in the PKT_COORD packet. The packet includes everything the server needs to know about the players’ movement state, including position, momentum, angles, and now also the forward and side move values. Ultimately, the position and momentum values should be used for spline-based movement paths, so that the movement of remote players will look smooth for the client. At the moment momentum is applied linearly, which will cause skipping when the momentum is changing (acceleration, deceleration and curving paths).
Also, I applied a couple of patches related to shared library linking on Linux, and 64-bit memory zone alignment. Thanks atupone and makovick!
Before I can move onto Hexen there are a couple of glitches in Heretic that need to be addressed. The Morph Ovum is broken because the chicken player state is not transmitted to the client, and the player shooting animation is not always shown — either because the server doesn’t play it, or because a delta about the state change won’t get transmitted for some reason.

danij:

(…)

skyjake:

Work continued on fixing the multiplayer. I managed to address two issues: player attack animation (broken in all games), and the behavior of the Heretic Morph Ovum. In the case of the attack animation, the problem was that the server was not sending all the state changes to the client. Apparently the original reason was to reduce the amount of transferred data, but in this case the implementation was resulting in an obvious malfunction, so the check was removed. At the moment the server will send mobj state changes to the clients when they are not part of the normal state animation sequence. With the Morph Ovum the problem was twofold: on serverside the players' class was not changed to Chicken, which meant that when the player attacked the mobj changed to the normal Corvus attack sequence. On clientside the psprites were never updated to reflect the Chicken class.
Also, there was some activity related to project infrastructure. As described in the previous post, the dengDevs blog was transitioned to Archive mode. The other change was that old builds (12+ weeks) will start getting purged from the builds repository. However, the build tags remain in the Git repository even after being purged, so each build can still be accessed (albeit only manually from source).
Unless I spot more issues with Heretic, I will shift my focus to Hexen where I believe there are quite a number of things broken. IIRC, Hexen was never fully updated during the last round of netgame related changes -- fortunately most of the required solutions already exist in the other two games.

danij:

(to be added)

skyjake:

The past week I was focusing squarely on Hexen's multiplayer. The issue I was wondering about last week (player getting stuck on objects) was rather straightforward to fix: one #ifdef __JHEXEN__ was on the wrong line causing the collision check to never clear an object's status for being on another object -- once the player landed on something, it was impossible to get rid of it. I also fixed a number of other problems, such as clientside polyobj rotation, immediate plane moves, and a mix-up with which player should be visible in the current viewport.

A couple of noteworthy points:

Extra care is necessary when dealing with the fixed-point and angle_t angles. It is quite easy to mess up their signs when converting between fixed and float and (de)serializing them for the deltas. We should have dedicated macros similar to FIX2FLT() for these operations.
The engine's global displayPlayer variable is no longer writable through the API. The games were expecting that it specifically defines the local console player's display, but as we now have multiple viewports, it is no longer useful fo that purpose. A new public API function was added for setting the displayplayer for a particular player (R_SetViewPortPlayer).

Hexen is now looking somewhat better for multiplayer, but I still need to hunt down more issues with it. I've yet to try a game with more than one client; I'm positive that some issues will become apparent when I do.

danij:

This week I found myself somewhat preoccupied with real life and typically, deng development suffered somewhat. There was some progress however - as set out by my last update, I've since externalized the built-in Patch Replacement definitions for jDoom's game menu and intermissions into jdoom.pk3 which, allowed me to further clean up the internal API.

I also fixed a couple of bugs in the font renderer and further improved the viewport management, adding the new hook HOOK_VIEWPORT_RESHAPE which the engine calls automatically upon viewport dimension change.

Over the coming week I hope to find rather more time for deng. The plan is to continue with the UI refactorings, beginning by replacing the remaining static menu init with dynamic page construction procedures.

последнюю, поддерживающую WinXP