fix github CI & stuff

workflows/build.yml:
- win32/64: install mingw32/64 gcc on msys (because the default
  gcc installed elsewhere seems to use ucrt, producing incompatible
  fp printf formats.)
tccgen.c:
- cleanup funcs: save any lvalues from return expressions.  Also use
  get_temp_local_var() which however was causing a problem on i386
  because its gfunc_call() removes the arguments from vstack and by
  that defeats the 'in-use'  tracking of get_temp_local_var().  Fixed by:
i386/arm/arm64/x86_64-gen.c:
- in gfunc_call(): save_regs before anything else, fixes
  problems seen in arm64/i386
tccpp.c:
- allow arm asm specific use of '#' in macros
libtcc.c:
- organize -M options, and:
tccpe.c:
- move the subsystem option parsing from libtcc.c
tccelf.c:
- improved error handling in tcc_load_ldscript()
lib/atomic.S:
- TCC_TARGET_... not defined when building the lib
- endbrNN security feature not supported by TCC
tests/tests2/136_atomic_gcc_style.c:
- never use standard assert() in tests

.github/workflows/build.yml

@@ -29,29 +29,41 @@ jobs:
     - name: make & test tcc (aarch64-osx)
       run: ./configure && make && make test -k
 
-  test-x86-win32:
-    runs-on: windows-2022
-    timeout-minutes: 6
+  test-x86_64-win32:
+    runs-on: windows-2025
+    timeout-minutes: 4
     steps:
     - uses: actions/checkout@v4
     - name: make & test tcc (x86_64-win32)
       shell: cmd
       run: |
+        echo ::group:: setup msys mingw64-gcc
         set MSYS2_PATH_TYPE=inherit
         set MSYSTEM=MINGW64
         set CHERE_INVOKING=yes
+        C:\msys64\usr\bin\bash -l -c "pacman -S --noconfirm mingw-w64-x86_64-gcc"
+        echo ::endgroup::
         C:\msys64\usr\bin\bash -l -c "./configure && make && make test -k"
+
+  test-i386-win32:
+    runs-on: windows-2025
+    timeout-minutes: 4
+    steps:
+    - uses: actions/checkout@v4
     - name: make & test tcc (i386-win32)
       shell: cmd
       run: |
+        echo ::group:: setup msys mingw32-gcc
         set MSYS2_PATH_TYPE=inherit
         set MSYSTEM=MINGW32
         set CHERE_INVOKING=yes
-        C:\msys64\usr\bin\bash -l -c "./configure && make clean all && make test -k"
+        C:\msys64\usr\bin\bash -l -c "pacman -S --noconfirm mingw-w64-i686-gcc"
+        echo ::endgroup::
+        C:\msys64\usr\bin\bash -l -c "./configure && make all && make test -k"
 
   test-armv7-linux:
     runs-on: ubuntu-22.04
-    timeout-minutes: 6
+    timeout-minutes: 8
     steps:
     - uses: actions/checkout@v4
     - uses: uraimo/run-on-arch-action@v3
@@ -69,14 +81,14 @@ jobs:
 
   test-aarch64-linux:
     runs-on: ubuntu-22.04
-    timeout-minutes: 6
+    timeout-minutes: 8
     steps:
     - uses: actions/checkout@v4
     - uses: uraimo/run-on-arch-action@v3
       name: make & test tcc (aarch64-linux)
       with:
         arch: aarch64
-        distro: ubuntu22.04
+        distro: ubuntu24.04
         githubToken: ${{ github.token }}
         install: |
           apt-get update -q -y
@@ -87,7 +99,7 @@ jobs:
 
   test-riscv64-linux:
     runs-on: ubuntu-22.04
-    timeout-minutes: 6
+    timeout-minutes: 8
     steps:
     - uses: actions/checkout@v4
     - uses: uraimo/run-on-arch-action@v3

Makefile

@@ -252,8 +252,11 @@ LDFLAGS += -g
 endif
 
 # convert "include/tccdefs.h" to "tccdefs_.h"
-%_.h : include/%.h conftest.c
-	$S$(CC) -DC2STR $(filter %.c,$^) -o c2str.exe && ./c2str.exe $< $@
+%_.h : include/%.h c2str.exe
+	$S./c2str.exe $< $@
+
+c2str.exe : conftest.c
+	$S$(CC) -DC2STR $< -o $@
 
 # target specific object rule
 $(X)%.o : %.c $(LIBTCC_INC)

arm-gen.c

@@ -1311,10 +1311,6 @@ again:
   if (++pass < 2)
     goto again;
 
-  /* Manually free remaining registers since next parameters are loaded
-   * manually, without the help of gv(int). */
-  save_regs(nb_args);
-
   if(todo) {
     o(0xE8BD0000|todo); /* pop {todo} */
     for(pplan = plan->clsplans[CORE_STRUCT_CLASS]; pplan; pplan = pplan->prev) {
@@ -1354,6 +1350,8 @@ void gfunc_call(int nb_args)
     gbound_args(nb_args);
 #endif
 
+  save_regs(nb_args + 1);
+
 #ifdef TCC_ARM_EABI
   if (float_abi == ARM_HARD_FLOAT) {
     variadic = (vtop[-nb_args].type.ref->f.func_type == FUNC_ELLIPSIS);

arm64-gen.c

@@ -997,6 +997,8 @@ ST_FUNC void gfunc_call(int nb_args)
     int variadic = (vtop[-nb_args].type.ref->f.func_type == FUNC_ELLIPSIS);
     int var_nb_arg = n_func_args(&vtop[-nb_args].type);
 
+    save_regs(nb_args + 1);
+
 #ifdef CONFIG_TCC_BCHECK
     if (tcc_state->do_bounds_check)
         gbound_args(nb_args);
@@ -1101,9 +1103,6 @@ ST_FUNC void gfunc_call(int nb_args)
             // value in floating-point registers
             if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
                 uint32_t j, sz, n = arm64_hfa(&vtop->type, &sz);
-                // save regs because struct may overwrite previous func call result
-                save_regs(0);
-
                 vtop->type.t = VT_PTR;
                 gaddrof();
                 gv(RC_R30);
@@ -1129,7 +1128,6 @@ ST_FUNC void gfunc_call(int nb_args)
             vswap();
     }
 
-    save_regs(0);
     arm64_gen_bl_or_b(0);
     --vtop;
     if (stack & 0xfff)

i386-gen.c

@@ -406,6 +406,8 @@ ST_FUNC void gfunc_call(int nb_args)
         gbound_args(nb_args);
 #endif
 
+    save_regs(nb_args + 1);
+
     args_size = 0;
     for(i = 0;i < nb_args; i++) {
         if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
@@ -463,7 +465,7 @@ ST_FUNC void gfunc_call(int nb_args)
         }
         vtop--;
     }
-    save_regs(0); /* save used temporary registers */
+
     func_sym = vtop->type.ref;
     func_call = func_sym->f.func_call;
     /* fast call case */

lib/atomic.S

@@ -23,8 +23,10 @@
 
 /* ---------------------------------------------- */
 #if defined __i386__
+
+#define endbr32
+
         .text
-        .align  2
 
         .global _(__atomic_load_1)
         .type   _(__atomic_load_1), %function
@@ -297,9 +299,11 @@ _(atomic_flag_clear_explicit):
 #endif //__i386__
 
 /* ---------------------------------------------- */
-#if defined __x86_64__ && !defined TCC_TARGET_PE
+#if defined __x86_64__ && !defined _WIN32
+
+#define endbr64
+
         .text
-        .align  2
 
         .global _(__atomic_load_1)
         .type   _(__atomic_load_1), %function
@@ -508,12 +512,11 @@ _(atomic_flag_clear_explicit):
         ret
 	.size   _(atomic_flag_clear_explicit), .-_(atomic_flag_clear_explicit)
 
-#endif //__x86_64__ && !TCC_TARGET_PE
+#endif //__x86_64__ && !_WIN32
 
 /* ---------------------------------------------- */
-#if defined __x86_64__ && defined TCC_TARGET_PE
+#if defined __x86_64__ && defined _WIN32
         .text
-        .align  2
 
         .global _(__atomic_load_1)
         .type   _(__atomic_load_1), %function
@@ -700,12 +703,11 @@ _(atomic_flag_clear_explicit):
         ret
 	.size   _(atomic_flag_clear_explicit), .-_(atomic_flag_clear_explicit)
 
-#endif //__x86_64__ && TCC_TARGET_PE
+#endif //__x86_64__ && _WIN32
 
 /* ---------------------------------------------- */
 #if defined __arm__
         .text
-        .align  2
 #ifndef __TINYC__
         .arch   armv6k
         .syntax unified
@@ -1496,7 +1498,6 @@ _(atomic_flag_clear_explicit):
 /* ---------------------------------------------- */
 #if defined __aarch64__
         .text
-        .align  2
 
         .global _(__atomic_load_1)
         .type   _(__atomic_load_1), %function
@@ -2115,7 +2116,6 @@ _(atomic_flag_clear_explicit):
 /* ---------------------------------------------- */
 #if defined __riscv
         .text
-        .align  2
 
         .global _(__atomic_load_1)
         .type   _(__atomic_load_1), %function

libtcc.c

@@ -1465,31 +1465,9 @@ static int tcc_set_linker(TCCState *s, const char *optarg)
         } else if (link_option(&o, "stack=")) {
             s->pe_stack_size = strtoul(o.arg, &end, 10);
         } else if (link_option(&o, "subsystem=")) {
-#if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
-            if (0==strcmp("native", o.arg)) {
-                s->pe_subsystem = 1;
-            } else if (0==strcmp("console", o.arg)) {
-                s->pe_subsystem = 3;
-            } else if (0==strcmp("gui", o.arg) || 0==strcmp("windows", o.arg)) {
-                s->pe_subsystem = 2;
-            } else if (0==strcmp("posix", o.arg)) {
-                s->pe_subsystem = 7;
-            } else if (0==strcmp("efiapp", o.arg)) {
-                s->pe_subsystem = 10;
-            } else if (0==strcmp("efiboot", o.arg)) {
-                s->pe_subsystem = 11;
-            } else if (0==strcmp("efiruntime", o.arg)) {
-                s->pe_subsystem = 12;
-            } else if (0==strcmp("efirom", o.arg)) {
-                s->pe_subsystem = 13;
-#elif defined(TCC_TARGET_ARM)
-            if (0==strcmp("wince", o.arg)) {
-                s->pe_subsystem = 9;
-#endif
-            } else
+            if (pe_setsubsy(s, o.arg) < 0)
                 goto err;
-#endif /* PE */
-#ifdef TCC_TARGET_MACHO
+#elif defined TCC_TARGET_MACHO
         } else if (link_option(&o, "all_load")) {
 	    s->filetype |= AFF_WHOLE_ARCHIVE;
         } else if (link_option(&o, "force_load=")) {
@@ -1653,11 +1631,10 @@ static const TCCOption tcc_options[] = {
     { "w", TCC_OPTION_w, 0 },
     { "E", TCC_OPTION_E, 0},
     { "M", TCC_OPTION_M, 0},
-    { "MD", TCC_OPTION_MD, 0},
-    { "MF", TCC_OPTION_MF, TCC_OPTION_HAS_ARG },
     { "MM", TCC_OPTION_MM, 0},
-    { "MMD,", TCC_OPTION_MMD, TCC_OPTION_HAS_ARG | TCC_OPTION_NOSEP },
-    { "MMD", TCC_OPTION_MMD, 0},
+    { "MD", TCC_OPTION_MD, TCC_OPTION_HAS_ARG | TCC_OPTION_NOSEP },
+    { "MMD", TCC_OPTION_MMD, TCC_OPTION_HAS_ARG | TCC_OPTION_NOSEP },
+    { "MF", TCC_OPTION_MF, TCC_OPTION_HAS_ARG },
     { "MP", TCC_OPTION_MP, 0},
     { "x", TCC_OPTION_x, TCC_OPTION_HAS_ARG },
     /* tcctools */
@@ -2075,6 +2052,7 @@ PUB_FUNC int tcc_parse_args(TCCState *s, int *pargc, char ***pargv)
         case TCC_OPTION_P:
             s->Pflag = atoi(optarg) + 1;
             break;
+
         case TCC_OPTION_M:
             s->include_sys_deps = 1;
             // fall through
@@ -2084,29 +2062,31 @@ PUB_FUNC int tcc_parse_args(TCCState *s, int *pargc, char ***pargv)
             if (!s->deps_outfile)
                 tcc_set_str(&s->deps_outfile, "-");
             break;
+        case TCC_OPTION_MD:
+            s->include_sys_deps = 1;
+            // fall through
         case TCC_OPTION_MMD:
             s->gen_deps = 1;
             /* usually, only "-MMD" is used */
             /* but the Linux Kernel uses "-MMD,depfile" */
-            if ((optarg) && (*optarg != '\0'))
-                tcc_set_str(&s->deps_outfile, optarg);
-            break;
-        case TCC_OPTION_MD:
-            s->gen_deps = 1;
-            s->include_sys_deps = 1;
-            break;
+            if (*optarg != ',')
+                break;
+            ++optarg;
+            // fall through
         case TCC_OPTION_MF:
             tcc_set_str(&s->deps_outfile, optarg);
             break;
         case TCC_OPTION_MP:
             s->gen_phony_deps = 1;
             break;
+
         case TCC_OPTION_dumpmachine:
             printf("%s\n", dumpmachine_str);
             exit(0);
         case TCC_OPTION_dumpversion:
             printf ("%s\n", TCC_VERSION);
             exit(0);
+
         case TCC_OPTION_x:
             x = 0;
             if (*optarg == 'c')

tcc.h

@@ -1790,6 +1790,7 @@ ST_FUNC void asm_clobber(uint8_t *clobber_regs, const char *str);
 ST_FUNC int pe_load_file(struct TCCState *s1, int fd, const char *filename);
 ST_FUNC int pe_output_file(TCCState * s1, const char *filename);
 ST_FUNC int pe_putimport(TCCState *s1, int dllindex, const char *name, addr_t value);
+ST_FUNC int pe_setsubsy(TCCState *s1, const char *arg);
 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
 #endif
 #ifdef TCC_TARGET_X86_64

tccelf.c

@@ -3937,7 +3937,7 @@ static int ld_add_file(TCCState *s1, const char filename[])
 {
     if (filename[0] == '-' && filename[1] == 'l')
         return tcc_add_library(s1, filename + 2);
-    {
+    if (CONFIG_SYSROOT[0] != '\0' || !IS_ABSPATH(filename)) {
         /* lookup via library paths */
         int ret = tcc_add_dll(s1, tcc_basename(filename), 0);
         if (ret != FILE_NOT_FOUND)
@@ -3981,19 +3981,19 @@ repeat:
         } else if (t != LD_TOK_NAME) {
             return tcc_error_noabort("unexpected token '%c'", t);
         } else if (!strcmp(filename, "AS_NEEDED")) {
-            ret = ld_add_file_list(s1, filename);
+            ret |= ld_add_file_list(s1, filename);
         } else if (c == 'I' || c == 'G' || c == 'A') {
-            ret = ld_add_file(s1, filename);
+            ret |= !!ld_add_file(s1, filename);
         }
-        if (ret)
-            return -1;
+        if (ret < 0)
+            return ret;
         t = ld_next(s1, filename, sizeof(filename));
         if (t == ',')
             t = ld_next(s1, filename, sizeof(filename));
     }
-    if (c == 'G' && new_undef_sym(s1, sym_offset))
+    if (c == 'G' && ret == 0 && new_undef_sym(s1, sym_offset))
         goto repeat;
-    return 0;
+    return ret;
 }
 
 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
@@ -4001,7 +4001,7 @@ repeat:
 ST_FUNC int tcc_load_ldscript(TCCState *s1, int fd)
 {
     char cmd[64];
-    int t, ret = FILE_NOT_RECOGNIZED;
+    int t, ret = 0, noscript = 1;
     unsigned char *text_ptr, *saved_ptr;
 
     saved_ptr = s1->ld_p;
@@ -4012,19 +4012,22 @@ ST_FUNC int tcc_load_ldscript(TCCState *s1, int fd)
             break;
         if (!strcmp(cmd, "INPUT") ||
             !strcmp(cmd, "GROUP")) {
-            ret = ld_add_file_list(s1, cmd);
+            ret |= ld_add_file_list(s1, cmd);
         } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
                    !strcmp(cmd, "TARGET")) {
             /* ignore some commands */
-            ret = ld_add_file_list(s1, cmd);
-        } else if (0 == ret) {
+            ret |= ld_add_file_list(s1, cmd);
+        } else if (noscript) {
+            ret = FILE_NOT_RECOGNIZED;
+        } else {
             ret = tcc_error_noabort("unexpected '%s'", cmd);
         }
-        if (ret)
+        if (ret < 0)
             break;
+        noscript = 0;
     }
     tcc_free(text_ptr);
     s1->ld_p = saved_ptr;
-    return ret;
+    return ret < 0 ? ret : -ret;
 }
 #endif /* !ELF_OBJ_ONLY */

tccgen.c

@@ -6839,15 +6839,15 @@ static void end_switch(void)
 /* ------------------------------------------------------------------------- */
 /* __attribute__((cleanup(fn))) */
 
-/* save SValue of symbol to local stack */
-static void save_cleanup_sym(Sym *s)
+/* protect symbol lvalues from further modification  */
+static void save_lvalues(void)
 {
     SValue *sv = vtop;
     while (sv >= vstack) {
-        if (sv->sym == s) {
+        if (sv->sym && (sv->r & VT_LVAL)) {
             int align, size = type_size(&sv->type, &align);
-            loc = (loc - size) & -align;
-            vset(&sv->type, VT_LOCAL | VT_LVAL, loc);
+            int r2, l = get_temp_local_var(size, align, &r2);
+            vset(&sv->type, VT_LOCAL | VT_LVAL, l), vtop->r2 = r2;
             vpushv(sv), *sv = vtop[-1], vstore(), --vtop;
         }
         --sv;
@@ -6860,7 +6860,7 @@ static void try_call_scope_cleanup(Sym *stop)
     for (; cls != stop; cls = cls->next) {
 	Sym *fs = cls->cleanup_func;
 	Sym *vs = cls->prev_tok;
-        save_cleanup_sym(vs);
+	save_lvalues();
 	vpushsym(&fs->type, fs);
 	vset(&vs->type, vs->r, vs->c);
 	vtop->sym = vs;

tccpe.c

@@ -1254,7 +1254,7 @@ static int pe_check_symbols(struct pe_info *pe)
         if (sym->st_shndx == SHN_UNDEF) {
             const char *name = (char*)symtab_section->link->data + sym->st_name;
             unsigned type = ELFW(ST_TYPE)(sym->st_info);
-            int imp_sym;
+            int imp_sym = 0;
             struct import_symbol *is;
 
             int _imp_, n;
@@ -1928,6 +1928,34 @@ static void pe_add_runtime(TCCState *s1, struct pe_info *pe)
     pe->type = pe_type;
 }
 
+ST_FUNC int pe_setsubsy(TCCState *s1, const char *arg)
+{
+    static const struct subsy { const char* p; int v; } x[] = {
+#if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
+        { "native", 1 },
+        { "gui", 2 },
+        { "windows", 2 },
+        { "console", 3 },
+        { "posix", 7 },
+        { "efiapp", 10 },
+        { "efiboot", 11 },
+        { "efiruntime", 12 },
+        { "efirom", 13 },
+#elif defined(TCC_TARGET_ARM)
+        { "wince", 9 },
+#endif
+        { 0, -1 }};
+    const struct subsy *y;
+    for (y = x;; ++y) {
+        if (!y->p)
+            return -1;
+        if (0 == strcmp(y->p, arg)) {
+            s1->pe_subsystem = y->v;
+            return 0;
+        }
+    }
+}
+
 static void pe_set_options(TCCState * s1, struct pe_info *pe)
 {
     if (PE_DLL == pe->type) {

tccpp.c

@@ -3076,6 +3076,11 @@ static int *macro_arg_subst(Sym **nested_list, const int *macro_str, Sym *args)
                 cval.str.size = tokcstr.size;
                 cval.str.data = tokcstr.data;
                 tok_str_add2(&str, TOK_PPSTR, &cval);
+#ifdef TCC_TARGET_ARM
+            } else if ((parse_flags & PARSE_FLAG_ASM_FILE) && t == TOK_PPNUM) {
+                /* for example: mov r1,#0 */
+                --macro_str, tok_str_add(&str, '#');
+#endif
             } else {
                 expect("macro parameter after '#'");
             }

tests/tests2/136_atomic_gcc_style.c

@@ -1,8 +1,11 @@
 #include <stdatomic.h>
 #include <stdbool.h>
-#include <assert.h>
 #include <stdio.h>
 
+// standard assert would popup a dialog box on windows
+#define assert(x) \
+    printf("assert \"%s\" : %s\n", #x, (x) ? "yes" : "no");
+
 int main() {
     // Test 1: Basic functionality of __atomic_store_n and __atomic_load_n
     {

tests/tests2/136_atomic_gcc_style.expect

@@ -1 +1,11 @@
+assert "loaded == 42" : yes
+assert "success" : yes
+assert "atomic_var == 200" : yes
+assert "expected == 100" : yes
+assert "!success" : yes
+assert "atomic_var == 100" : yes
+assert "expected == 100" : yes
+assert "atomic_var == 60" : yes
+assert "loaded_ptr == NULL" : yes
+assert "__atomic_load_n(&atomic_var, __ATOMIC_RELAXED) == 10" : yes
 All atomic tests passed!

x86_64-gen.c

@@ -809,6 +809,8 @@ void gfunc_call(int nb_args)
         gbound_args(nb_args);
 #endif
 
+    save_regs(nb_args);
+
     args_size = (nb_args < REGN ? REGN : nb_args) * PTR_SIZE;
     arg = nb_args;
 
@@ -906,7 +908,7 @@ void gfunc_call(int nb_args)
         }
         vtop--;
     }
-    save_regs(0);
+
     /* Copy R10 and R11 into RCX and RDX, respectively */
     if (nb_args > 0) {
         o(0xd1894c); /* mov %r10, %rcx */
@@ -1249,6 +1251,8 @@ void gfunc_call(int nb_args)
         gbound_args(nb_args);
 #endif
 
+    save_regs(nb_args);
+
     /* calculate the number of integer/float register arguments, remember
        arguments to be passed via stack (in onstack[]), and also remember
        if we have to align the stack pointer to 16 (onstack[i] == 2).  Needs
@@ -1364,9 +1368,6 @@ void gfunc_call(int nb_args)
 
     tcc_free(onstack);
 
-    /* XXX This should be superfluous.  */
-    save_regs(0); /* save used temporary registers */
-
     /* then, we prepare register passing arguments.
        Note that we cannot set RDX and RCX in this loop because gv()
        may break these temporary registers. Let's use R10 and R11
@@ -1416,12 +1417,6 @@ void gfunc_call(int nb_args)
     assert(gen_reg == 0);
     assert(sse_reg == 0);
 
-    /* We shouldn't have many operands on the stack anymore, but the
-       call address itself is still there, and it might be in %eax
-       (or edx/ecx) currently, which the below writes would clobber.
-       So evict all remaining operands here.  */
-    save_regs(0);
-
     /* Copy R10 and R11 into RDX and RCX, respectively */
     if (nb_reg_args > 2) {
         o(0xd2894c); /* mov %r10, %rdx */