fix: correct check of BOM using open(2)

this emulates the implementation of _tcc_open which uses also open with a
O_BINARY flag which is zero on POSIX. Another approach may be to change the tcc
code, but I prefer to not touch that when possible.

src/file.c

@@ -23,6 +23,9 @@
 #include <string.h>
 #include <errno.h>
 #include <time.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <inttypes.h>
 
 #include <ftw.h> // _GNU_SOURCE
 
@@ -36,27 +39,35 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <dirent.h>
+#ifndef O_BINARY
+# define O_BINARY 0
+#endif
 #endif
 extern void _err(const char *fmt, ...);
 
 // from exec-headers.c
 extern bool gen_exec_headers(char *tmpdir);
 
 int detect_bom(const char *filename) {
-    FILE *file = fopen(filename, "rb");
-    if (!file) return -1;
-    char bom[3];
-    fread(bom, 1, 3, file);
-    fclose(file);
-    if (bom[0] == 0xFF && bom[1] == 0xFE) {
-      return 1; // UTF-16 LE
-    } else if (bom[0] == 0xFE && bom[1] == 0xFF) {
-      return 2; // UTF-16 BE
-    } else if (bom[0] == 0xEF && bom[1] == 0xBB && bom[2] == 0xBF) {
-      return 3; // UTF-8
-    } else {
-      return 0; // No BOM
-    }
+	uint8_t bom[3];
+	int res;
+	int fd = open(filename, O_RDONLY | O_BINARY);
+	res = read(fd,bom,3);
+	if (res!=3) {
+		_err("read error on %s: %s",filename, strerror(errno));
+		return -1;
+	}
+	close(fd);
+	// _err("%s bom: %x %x %x",filename,bom[0],bom[1],bom[2]);
+	if (bom[0] == 0xFF && bom[1] == 0xFE) {
+		return 1; // UTF-16 LE
+	} else if (bom[0] == 0xFE && bom[1] == 0xFF) {
+		return 2; // UTF-16 BE
+	} else if (bom[0] == 0xEF && bom[1] == 0xBB && bom[2] == 0xBF) {
+		return 3; // UTF-8
+	} else {
+		return 0; // No BOM
+	}
 }
 
 bool append_path(char **stored_path, const char *new_path) {

test/windows.bats

@@ -8,6 +8,8 @@ load bats_setup
 }
 
 @test "BOM source file UTF8" {
-      run ${CJIT} test/hello-bom-utf8.c
+      run ${CJIT} -q test/hello-bom-utf8.c
       assert_failure
+      assert_line 'UTF BOM detected in file: test/hello-bom-utf8.c'
+      assert_line 'Encoding is not yet supported, execution aborted.'
 }