Keep narrow directory name around some more

What this did was

1. Find directory
2. Turn name into wcstring and return it
3. Turn name back into string for some operations

Instead, let's unglue the wcstringing from this, return the narrow
string and then widen it when we need.

src/wildcard.cpp

@@ -736,16 +736,16 @@ void wildcard_expander_t::expand_intermediate_segment(const wcstring &base_dir,
                                                       const wcstring &wc_segment,
                                                       const wchar_t *wc_remainder,
                                                       const wcstring &prefix) {
-    wcstring name_str;
+    std::string narrow;
     int dir_fd = dirfd(base_dir_fp);
-    while (!interrupted_or_overflowed() && wreaddir_for_dirs(base_dir_fp, &name_str)) {
+    while (!interrupted_or_overflowed() && readdir_for_dirs(base_dir_fp, &narrow)) {
+        wcstring name_str = str2wcstring(narrow);
         // Note that it's critical we ignore leading dots here, else we may descend into . and ..
         if (!wildcard_match(name_str, wc_segment, true)) {
             // Doesn't match the wildcard for this segment, skip it.
             continue;
         }
 
-        std::string narrow = wcs2string(name_str);
         struct stat buf;
         if (0 != fstatat(dir_fd, narrow.c_str(), &buf, 0) || !S_ISDIR(buf.st_mode)) {
             // We either can't stat it, or we did but it's not a directory.
@@ -776,18 +776,19 @@ void wildcard_expander_t::expand_literal_intermediate_segment_with_fuzz(const wc
                                                                         const wchar_t *wc_remainder,
                                                                         const wcstring &prefix) {
     // This only works with tab completions. Ordinary wildcard expansion should never go fuzzy.
-    wcstring name_str;
+    std::string narrow;
 
     // Mark that we are fuzzy for the duration of this function
     const scoped_push<bool> scoped_fuzzy(&this->has_fuzzy_ancestor, true);
 
     int dir_fd = dirfd(base_dir_fp);
-    while (!interrupted_or_overflowed() && wreaddir_for_dirs(base_dir_fp, &name_str)) {
+    while (!interrupted_or_overflowed() && readdir_for_dirs(base_dir_fp, &narrow)) {
         // Don't bother with . and ..
-        if (name_str == L"." || name_str == L"..") {
+        if (narrow == "." || narrow == "..") {
             continue;
         }
 
+        wcstring name_str = str2wcstring(narrow);
         // Skip cases that don't match or match exactly. The match-exactly case was handled directly
         // in expand().
         const maybe_t<string_fuzzy_match_t> match = string_fuzzy_match_string(wc_segment, name_str);

src/wutil.cpp

@@ -94,7 +94,7 @@ bool wreaddir(DIR *dir, wcstring &out_name) {
     return true;
 }
 
-bool wreaddir_for_dirs(DIR *dir, wcstring *out_name) {
+bool readdir_for_dirs(DIR *dir, std::string *out_name) {
     struct dirent *result = nullptr;
     while (!result) {
         result = readdir(dir);
@@ -118,7 +118,7 @@ bool wreaddir_for_dirs(DIR *dir, wcstring *out_name) {
     }
 
     if (result && out_name) {
-        *out_name = str2wcstring(result->d_name);
+        *out_name = result->d_name;
     }
     return result != nullptr;
 }

src/wutil.h

@@ -71,7 +71,7 @@ bool wreaddir_resolving(DIR *dir, const std::wstring &dir_path, wcstring &out_na
 /// (i.e. the file is a symlink), then return it. Note that this does not guarantee that everything
 /// returned is a directory, it's just an optimization for cases where we would check for
 /// directories anyways.
-bool wreaddir_for_dirs(DIR *dir, wcstring *out_name);
+bool readdir_for_dirs(DIR *dir, std::string *out_name);
 
 /// Wide character version of dirname().
 std::wstring wdirname(std::wstring path);