Repository navigation
Fix turret locks, red-alert damage and subsystem flags after a ship class change - #7923
Goober5000 wants to merge 4 commits into
Conversation
dc3ab31 to
b544d0f
Compare
b544d0f to
e6e33e5
Compare
dustinspace217
left a comment
There was a problem hiding this comment.
I built this branch and its merge base (a981ff0) and ran them side by side on retail FS2 with MediaVPs 5.0.2: Release builds on OpenGL and Vulkan, plus Debug builds for the red-alert runs, using a small test mod with three missions.
- Turret lock at arrival: a lock-all-turrets GTB Ursa switched to a Medusa before it arrived came in with its turret free and firing on master. On this branch it's locked.
- Mid-mission
change-ship-class: a lock-all-turrets Leviathan with one turret freed, one beam-freed, its sensors untargetable and its engine at 40% was changed to a Mentu. Master came out with every turret free and the beam-free and untargetable settings gone. This branch kept all of them, and the engine stayed at 40% on both. A Mentu changed to a Leviathan still picked up the Leviathan's own untargetable armor plates, so the old class's table settings don't carry over. - Red alert, with "$Red-alert applies to delayed ships" turned on and the carried ships arriving 3 seconds into the second mission: on master the Ursa's communications damage went onto its turret entry and communications came back at full strength, the Lilith's sensor damage went onto its engine, and the Orion's fighterbay came back with NaN hitpoints. On this branch all three came through right, and the Debug build logged nothing new.
I've left a few notes inline. None of them block.
The flag was applied only inside the loop over the parse object's subsystem entries. FRED writes an entry for every turret of the ship's original class, but after a class change (ship select, change-ship-class on a ship that hasn't arrived, red alert) the new class's turrets that have no entry started unlocked. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Red alert restored a ship's subsystem damage by position, assuming that the parse object's first entry is the Pilot and the class's subsystems follow in order. FRED only writes turrets plus subsystems that are damaged or carry cargo, so any gap shifted the damage onto the wrong subsystems and created duplicate entries, even for unmodified FRED missions; for example, a GTB Ursa's communications damage landed on its turret. Look each subsystem up by name instead, creating an entry if there is none, as parse_copy_damage does. Also stop at the class's subsystem count, and skip subsystems with no hitpoints, which produced a NaN damage percentage. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
change_ship_type rebuilds the ship's subsystems, which resets every subsystem and turret weapon flag to what the new class's table gives it. Only hit percentages, blown-off state and AI references were carried over, so turret locks and beam-free state, including the lock applied by lock-all-turrets, and flags set by the ship-subsys sexps were lost. This affects wings that change class in ship select as well as change-ship-class during a mission. The new subsystems now start from the same defaults a newly created ship would have, including the lock from lock-all-turrets. Then, for each subsystem the new class shares by name, the runtime flags that differ from the old subsystem's defaults are carried over, so changes made since the ship was created survive but the old class's own table settings don't override the new class's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…empty list insert_subsys_status increments the parse object's subsystem count and then bumps the base index of every parse object at or after the new entry. When the parse object had no subsystems, its own base index equals the new entry's, so it was bumped as well: its one entry then became the next ship's first entry, or lay past the end of the array. The removed red-alert Assert was the only thing that caught this. FRED always writes a Pilot entry, so this needs a hand-edited mission, but it affected both red alert and parse_copy_damage. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
e6e33e5 to
a6b03fc
Compare
dustinspace217
left a comment
There was a problem hiding this comment.
Thanks for the updates. All four of my 11-10 notes are addressed, and I'm still approving.
I built the new head (a6b03fc) and master 51041cf as Release builds and ran them on retail FS2 with MediaVPs 5.0.2, in SM1-01 on OpenGL, and the branch again on Vulkan. I created a Leviathan, ran turret-lock-all, freed Turret01, made its sensors untargetable, and changed it to a Mentu. On this branch Turret02 to Turret08 stayed locked, Turret01 stayed free, the Mentu-only turrets came up free and the sensors stayed untargetable, the same on both renderers. Master came out with every turret free and the sensors targetable.
I also checked whether alter-ship-flag lock-all-turrets on a ship that has already arrived could change the baseline the class change compares against. It can't, since the flag isn't in Ship_flag_names, and in the run neither the sexp nor Lua setFlag changed it.
Smaller:
- code/mission/missionparse.cpp:8890: I didn't rerun the hand-edited empty-list red-alert case on this head, because I couldn't load a hand-edited mission in this run, but the skip reads right to me.
- code/ship/ship.cpp:12252:
ship-subsys-guardian-thresholdsets a field rather than a flag, sosubsys_setzeroes it at ship.cpp:7950, the restore loop doesn't bring it back, and a guarded engine can be destroyed afterchange-ship-class. Carryingsubsys_guardian_thresholdfor matched subsystems would fix it, or the description could drop "anything set by theship-subsys-*sexps" and list it under what isn't carried over. - code/ship/ship.cpp:12230: in multiplayer
turret-freeonly changes the server, so after a class change on a lock-all ship a freed shared turret stays free on the server and locked on the client, where aiturret.cpp:2198 then skips local tracking. Master's class change happened to reset both sides, and syncing the free and lock sexps is a separate fix, so a line in the description's limits would cover it here. - code/mission/missionparse.cpp:670 and ship_flags.h:104: "on a class change" can read as every turret relocking, while shared turrets a sexp freed stay free. This builds on my earlier help-text note; maybe "after a class change, turrets the new class adds also start locked".
- code/ship/ship.cpp:12113: this corrects my 11-10 approval, since the Mentu has no armor plates, so that run only showed new subsystems getting their table defaults, not the old class's table flags staying behind. The code compares against the old class's table as intended; I just haven't run two classes whose tables disagree on a shared subsystem.
Follow-up to #7910. While reviewing it, I found three older bugs in what happens to a ship's subsystems when its class changes. This PR fixes them.
Stacked on #7922, which restructures thesubsys_setflag setup that the third commit reuses. In draft until that PR merges. Only the last three commits belong to this PR.Apply lock-all-turrets to every turret when a ship is created
lock-all-turretswas applied inside the loop over the parse object's+Subsystem:entries, so it reached only turrets that have an entry. FRED writes one for every turret of the ship's original class, so it normally works. But after a class change (ship select,change-ship-classbefore arrival, red alert), any turret of the new class without an entry started unlocked. The lock is now applied by walking the ship's actual turrets.Match red-alert subsystem damage to the mission's entries by name
Red alert restored subsystem damage on unarrived ships by position. It assumed entry 0 is the Pilot and the class's subsystems follow in order. FRED only writes turrets plus subsystems that are damaged or carry cargo, so any gap shifted the damage onto the wrong subsystems and created duplicate entries. This happened even with unmodified FRED missions; for example:
Each subsystem is now looked up by name, and an entry is created if it's missing, the same way
parse_copy_damagedoes. The loop also stops at the class's subsystem count, and skips subsystems with no hitpoints, which used to produce a NaN damage percentage.Keep runtime subsystem state when a ship's class changes
change_ship_typerebuilds the subsystems from the new class's table, so every subsystem and turret weapon flag was reset. Only hit percentages, blown-off state and AI references were carried over. Turret locks and beam-free state were lost, including the lock fromlock-all-turrets, along with anything set by theship-subsys-*sexps. This affected wings that change class in ship select as well aschange-ship-classmid-mission.The new subsystems now start from the defaults a newly created ship would have, including the
lock-all-turretslock. Then, for each subsystem the new class shares by name, the runtime flags that differ from the old subsystem's defaults are carried over:Turret_LockUntargetableBeam_FreeNo_SS_targetingTagged_OnlyNo_replaceNo_live_debrisVanishedMissiles_ignore_if_deadEach class change compares against the class the ship has at that moment, so a flag that the mission changed and that matches the intermediate class's table can revert on a later class change. This situation needs per-subsystem table flags that differ between classes and two or more class changes, so it's a rare case that's deliberately not handled.
Changes made since the ship was created survive, but settings that only came from the old class's own table don't override the new class's. With
lock-all-turrets, a turret that is still locked stays locked, and one freed by a sexp stays free. A turret that exists only on the new class starts locked, as it would on a newly created ship.Forced turret targets, rotation and translation locks, cargo scan state, subsystem cargo and turret weapons are still not carried over.
🤖 Generated with Claude Code