Size-optimise cvar value cleanup

Put all the cvars together contiguously; then we don't need to generate
lots of individual calls to extfree() and can instead simply iterate a
fixed-sized array.

5 files changed, 47 insertions, 22 deletions

diff --git a/src/build/gluegen.c b/src/build/gluegen.c
index b2887a9..3e04db8 100644
--- a/src/build/gluegen.c
+++ b/src/build/gluegen.c
@@ -867,22 +867,6 @@ F( "	if (has_%.*s) _feat_end_%.*s();",
 	}
 _( "}")
 _( "")
-_( "static inline void freevars() {")
-	for (int i = 1; i < ncvars; ++i) {
-F( "	extfree(con_getvarcommon(%.*s)->strval);",
-		cvar_names[i].len, cvar_names[i].s)
-	}
-_( "}")
-_( "")
-_( "static inline void shuntvars() {")
-_( "#ifdef _WIN32")
-	for (int i = 1; i < ncvars; ++i) {
-F( "	__builtin_memmove(&%.*s->v1, &%.*s->v2, ",
-		cvar_names[i].len, cvar_names[i].s, cvar_names[i].len, cvar_names[i].s)
-_( "			sizeof(struct con_var_common));")
-	}
-_( "#endif")
-_( "}")
 	for (int i = 1; i < nevents; ++i) {
 		const char *prefix = event_predicateflags[i] ?
 				"bool CHECK_" : "void EMIT_";
diff --git a/src/con_.S b/src/con_.S
index 0c6ec2d..a9a7447 100644
--- a/src/con_.S
+++ b/src/con_.S
@@ -42,4 +42,24 @@ ASM_MANGLE(_con_colourmsg):
 	.byte 0x25
 	.long ASM_MANGLE(_con_colourmsgf)
 
+// The DEF_CVAR macros put stuff in .rdata$y/.rodata.cvars1. Sandwiching that
+// between empty x/0 and z/2 sections gives us section start/end pointers to
+// iterate the cvars, which is used for freeing the values on unload. Relies
+// on the section merging not adding extra padding, which is fine with the
+// linker + settings we use in practice.
+#ifdef _WIN32
+.section .rdata$x, "rw"
+#else
+.section .rodata.cvars0, "rw"
+#endif
+.globl ASM_MANGLE(_con_vars_start)
+ASM_MANGLE(_con_vars_start):
+#ifdef _WIN32
+.section .rdata$z, "rw"
+#else
+.section .rodata.cvars2, "rw"
+#endif
+.globl ASM_MANGLE(_con_vars_end)
+ASM_MANGLE(_con_vars_end):
+
 // vi: sw=4 ts=4 noet tw=80 cc=80
diff --git a/src/con_.c b/src/con_.c
index 54574bc..1d7d06e 100644
--- a/src/con_.c
+++ b/src/con_.c
@@ -22,6 +22,7 @@
 #include <string.h>
 
 #include "abi.h"
+#include "asm.h"
 #include "chunklets/x86.h"
 #include "con_.h"
 #include "engineapi.h" // for factories and rgba - XXX: is this a bit circular?
@@ -503,6 +504,16 @@ static inline void badver() {
 	con_warn("sst: error: this engine version is not yet supported\n");
 }
 
+extern struct con_var _con_vars_start, _con_vars_end; // con_.S
+
+static inline void shuntvars() {
+#ifdef _WIN32
+	for (struct con_var *v = &_con_vars_start; v != &_con_vars_end; ++v) {
+		__builtin_memmove(&v->v1, &v->v2, sizeof(struct con_var_common));
+	}
+#endif
+}
+
 bool con_detect(int pluginver) {
 	if (_con_iface = factory_engine("VEngineCvar007", 0)) {
 		// GENIUS HACK (BUT STILL BAD): Portal 2 has everything in ICvar shifted
@@ -574,6 +585,7 @@ bool con_detect(int pluginver) {
 	}
 	if (_con_iface = factory_engine("VEngineCvar003", 0)) {
 #ifdef _WIN32 // there's no OE on linux!
+		shuntvars(); // just doing this as early as possible atm, it's fine
 		_gametype_tag |= _gametype_tag_OE;
 		// for deletion/unlinking on unload, we need an indirect linked list
 		// pointer. calling GetCommands gives us a direct pointer. so we have to
@@ -733,6 +745,12 @@ void con_init() {
 	*pi++ = (void *)&GetSplitScreenPlayerSlot;
 }
 
+static inline void freevars() {
+	for (struct con_var *v = &_con_vars_start; v != &_con_vars_end; ++v) {
+		extfree(con_getvarcommon(v)->strval);
+	}
+}
+
 void con_disconnect() {
 #ifdef _WIN32
 	if (linkedlist) {
@@ -749,10 +767,11 @@ void con_disconnect() {
 				pp = next;
 			}
 		}
-		return;
+		goto f;
 	}
 #endif
 	UnregisterConCommands(_con_iface, dllid);
+f:	freevars();
 }
 
 struct con_var *con_findvar(const char *name) {
diff --git a/src/con_.h b/src/con_.h
index 5de5608..9202176 100644
--- a/src/con_.h
+++ b/src/con_.h
@@ -273,7 +273,14 @@ extern struct _con_vtab_iconvar_wrap {
 } _con_vtab_iconvar_wrap;
 #define _con_vtab_iconvar _con_vtab_iconvar_wrap.vtable
 
+#ifdef _WIN32
+#define _CON_VAR_SEG ".rdata$y"
+#else
+#define _CON_VAR_SEG ".rodata.cvars1"
+#endif
+
 #define _DEF_CVAR(name_, desc, value, hasmin_, min, hasmax_, max, flags_) \
+	__attribute__((section(_CON_VAR_SEG))) \
 	static struct con_var _cvar_##name_ = { \
 		.base = { \
 			.vtable = _con_vtab_var, \
diff --git a/src/sst.c b/src/sst.c
index e732073..bbcae79 100644
--- a/src/sst.c
+++ b/src/sst.c
@@ -14,8 +14,6 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-#include <string.h>
-
 #ifdef _WIN32
 #include <Windows.h>
 #include <shlwapi.h>
@@ -28,7 +26,6 @@
 #include "engineapi.h"
 #include "errmsg.h"
 #include "event.h"
-#include "extmalloc.h" // for freevars() in generated code
 #include "feature.h"
 #include "fixes.h"
 #include "gamedata.h"
@@ -543,7 +540,6 @@ static opt_small void hook_plugin_unload_cbv2(struct con_cmdargs *args) {
 static inline bool do_load(ifacefactory enginef, ifacefactory serverf) {
 	factory_engine = enginef; factory_server = serverf;
 	if_cold (!engineapi_init(ifacever)) return false;
-	if (GAMETYPE_MATCHES(OE)) shuntvars(); // see also comment in con_detect()
 	const void **p = vtable_firstdiff;
 	if (GAMETYPE_MATCHES(Portal2)) *p++ = (void *)&nop_p_v; // ClientFullyConnect
 	*p++ = (void *)&nop_p_v;		  // ClientDisconnect
@@ -598,7 +594,6 @@ static inline void do_unload() {
 	}
 	endfeatures();
 	con_disconnect();
-	freevars();
 }
 
 static bool VCALLCONV Load(void *this, ifacefactory enginef,