Ruby: playing with require and load

0001-make-load-and-require-use-file-expand_path-and-file-realpath.diff

diff --git a/file.c b/file.c
index 3107895..8bb0839 100644
--- a/file.c
+++ b/file.c
@@ -5117,7 +5117,7 @@ rb_find_file_ext_safe(VALUE *filep, const char *const *ext, int safe_level)
     if (!ext[0]) return 0;
 
     if (f[0] == '~') {
-	fname = file_expand_path_1(fname);
+	fname = rb_funcall(rb_cFile, rb_intern("expand_path"), 1, fname);
 	if (safe_level >= 1 && OBJ_TAINTED(fname)) {
 	    rb_raise(rb_eSecurityError, "loading from unsafe file %s", f);
 	}
@@ -5130,7 +5130,7 @@ rb_find_file_ext_safe(VALUE *filep, const char *const *ext, int safe_level)
 	if (safe_level >= 1 && !fpath_check(fname)) {
 	    rb_raise(rb_eSecurityError, "loading from unsafe path %s", f);
 	}
-	if (!expanded) fname = file_expand_path_1(fname);
+	if (!expanded) fname = rb_funcall(rb_cFile, rb_intern("expand_path"), 1, fname);
 	fnlen = RSTRING_LEN(fname);
 	for (i=0; ext[i]; i++) {
 	    rb_str_cat2(fname, ext[i]);
@@ -5151,7 +5151,6 @@ rb_find_file_ext_safe(VALUE *filep, const char *const *ext, int safe_level)
     if (!load_path) return 0;
 
     fname = rb_str_dup(*filep);
-    RBASIC(fname)->klass = 0;
     fnlen = RSTRING_LEN(fname);
     tmp = rb_str_tmp_new(MAXPATHLEN + 2);
     for (j=0; ext[j]; j++) {
@@ -5161,14 +5160,14 @@ rb_find_file_ext_safe(VALUE *filep, const char *const *ext, int safe_level)
 
 	    RB_GC_GUARD(str) = rb_get_path_check(str, safe_level);
 	    if (RSTRING_LEN(str) == 0) continue;
-	    file_expand_path(fname, str, 0, tmp);
+	    tmp = rb_funcall(rb_cFile, rb_intern("expand_path"), 2, fname, str);
 	    if (file_load_ok(RSTRING_PTR(tmp))) {
 		*filep = copy_path_class(tmp, *filep);
 		return (int)(j+1);
 	    }
 	    FL_UNSET(tmp, FL_TAINT | FL_UNTRUSTED);
 	}
-	rb_str_set_len(fname, fnlen);
+	rb_str_resize(fname, fnlen);
     }
     RB_GC_GUARD(load_path);
     return 0;
@@ -5188,7 +5187,7 @@ rb_find_file_safe(VALUE path, int safe_level)
     int expanded = 0;
 
     if (f[0] == '~') {
-	tmp = file_expand_path_1(path);
+	tmp = rb_funcall(rb_cFile, rb_intern("expand_path"), 1, path);
 	if (safe_level >= 1 && OBJ_TAINTED(tmp)) {
 	    rb_raise(rb_eSecurityError, "loading from unsafe file %s", f);
 	}
@@ -5203,7 +5202,7 @@ rb_find_file_safe(VALUE path, int safe_level)
 	}
 	if (!file_load_ok(f)) return 0;
 	if (!expanded)
-	    path = copy_path_class(file_expand_path_1(path), path);
+	    path = copy_path_class(rb_funcall(rb_cFile, rb_intern("expand_path"), 1, path), path);
 	return path;
     }
 
@@ -5220,7 +5219,7 @@ rb_find_file_safe(VALUE path, int safe_level)
 	    VALUE str = RARRAY_PTR(load_path)[i];
 	    RB_GC_GUARD(str) = rb_get_path_check(str, safe_level);
 	    if (RSTRING_LEN(str) > 0) {
-		file_expand_path(path, str, 0, tmp);
+		tmp = rb_funcall(rb_cFile, rb_intern("expand_path"), 2, path, str);
 		f = RSTRING_PTR(tmp);
 		if (file_load_ok(f)) goto found;
 	    }
diff --git a/load.c b/load.c
index 9ed386d..3bd0037 100644
--- a/load.c
+++ b/load.c
@@ -43,7 +43,7 @@ rb_get_expanded_load_path(void)
 
     ary = rb_ary_new2(RARRAY_LEN(load_path));
     for (i = 0; i < RARRAY_LEN(load_path); ++i) {
-	VALUE path = rb_file_expand_path(RARRAY_PTR(load_path)[i], Qnil);
+	VALUE path = rb_funcall(rb_cFile, rb_intern("expand_path"), 1, RARRAY_PTR(load_path)[i]);
 	rb_str_freeze(path);
 	rb_ary_push(ary, path);
     }
@@ -233,7 +233,7 @@ rb_feature_provided(const char *feature, const char **loading)
 
     if (*feature == '.' &&
 	(feature[1] == '/' || strncmp(feature+1, "./", 2) == 0)) {
-	fullpath = rb_file_expand_path(rb_str_new2(feature), Qnil);
+	fullpath = rb_funcall(rb_cFile, rb_intern("expand_path"), 1, rb_str_new2(feature));
 	feature = RSTRING_PTR(fullpath);
     }
     if (ext && !strchr(ext, '/')) {
@@ -305,7 +305,7 @@ rb_load_internal(VALUE fname, int wrap)
 	th->mild_compile_error++;
 	node = (NODE *)rb_load_file(RSTRING_PTR(fname));
 	loaded = TRUE;
-	iseq = rb_iseq_new_top(node, rb_str_new2("<top (required)>"), fname, rb_realpath_internal(Qnil, fname, 1), Qfalse);
+	iseq = rb_iseq_new_top(node, rb_str_new2("<top (required)>"), fname, rb_funcall(rb_cFile, rb_intern("realpath"), 1, fname), Qfalse);
 	th->mild_compile_error--;
 	rb_iseq_eval(iseq);
     }

console.log

V:\a\b\c\d\e>set RUBYOPT
RUBYOPT=-rV:/ruby-perf/replace

V:\a\b\c\d\e>dir /B
foo.rb

V:\a\b\c\d\e>ruby --disable-gems -I. -e "require 'foo'; File.results"
$LOAD_PATH: 9
$LOADED_FEATURES: 9
expanded: 82
absolute: 0
realpath: 1

replace.rb

$expanded = []
$absolute = []
$realpath = []

class File < IO
  class << self
    alias_method :orig_expand_path, :expand_path

    def expand_path(file_name, dir_string = nil)
      $expanded << "expand_path: #{file_name.inspect}, #{dir_string.inspect}"
      orig_expand_path file_name, dir_string
    end

    alias_method :orig_absolute_path, :absolute_path

    def absolute_path(file_name, dir_string = nil)
      $absolute << "absolute_path: #{file_name.inspect}, #{dir_string.inspect}"
      orig_absolute_path(file_name, dir_string)
    end

    alias_method :orig_realpath, :realpath

    def realpath(pathname, dir_string = nil)
      $realpath << "realpath: #{pathname.inspect}, #{dir_string.inspect}"
      orig_realpath pathname, dir_string
    end

    def results
      puts "$LOAD_PATH: #{$LOAD_PATH.size}"
      puts "$LOADED_FEATURES: #{$LOADED_FEATURES.size}"
      puts "expanded: #{$expanded.size}"
      puts "absolute: #{$absolute.size}"
      puts "realpath: #{$realpath.size}"
    end
  end
end

Another interesting detail:

V:\a\b\c\d\e>ruby --disable-gems -I. -e "puts $LOAD_PATH"
V:/a/b/c/d/e
V:/sandbox/ruby19_mingw/lib/ruby/site_ruby/1.9.1
V:/sandbox/ruby19_mingw/lib/ruby/site_ruby/1.9.1/i386-msvcrt
V:/sandbox/ruby19_mingw/lib/ruby/site_ruby
V:/sandbox/ruby19_mingw/lib/ruby/vendor_ruby/1.9.1
V:/sandbox/ruby19_mingw/lib/ruby/vendor_ruby/1.9.1/i386-msvcrt
V:/sandbox/ruby19_mingw/lib/ruby/vendor_ruby
V:/sandbox/ruby19_mingw/lib/ruby/1.9.1
V:/sandbox/ruby19_mingw/lib/ruby/1.9.1/i386-mingw32

Usage of -I in the command line results in a expanded path before it is added to $LOAD_PATH, not so when using unshift (btw, you shouldn't)