mirror of
https://github.com/ruby/ruby.git
synced 2022-11-09 12:17:21 -05:00
5e22f873ed
Saves comitters' daily life by avoid #include-ing everything from internal.h to make each file do so instead. This would significantly speed up incremental builds. We take the following inclusion order in this changeset: 1. "ruby/config.h", where _GNU_SOURCE is defined (must be the very first thing among everything). 2. RUBY_EXTCONF_H if any. 3. Standard C headers, sorted alphabetically. 4. Other system headers, maybe guarded by #ifdef 5. Everything else, sorted alphabetically. Exceptions are those win32-related headers, which tend not be self- containing (headers have inclusion order dependencies).
727 lines
19 KiB
C
727 lines
19 KiB
C
#if defined(__MINGW32__)
|
|
/* before stdio.h in ruby/define.h */
|
|
# define MINGW_HAS_SECURE_API 1
|
|
#endif
|
|
#include "ruby/ruby.h"
|
|
#include "ruby/encoding.h"
|
|
#include "internal.h"
|
|
#include "internal/error.h"
|
|
#include <winbase.h>
|
|
#include <wchar.h>
|
|
#include <shlwapi.h>
|
|
#include "win32/file.h"
|
|
|
|
#ifndef INVALID_FILE_ATTRIBUTES
|
|
# define INVALID_FILE_ATTRIBUTES ((DWORD)-1)
|
|
#endif
|
|
|
|
/* cache 'encoding name' => 'code page' into a hash */
|
|
static struct code_page_table {
|
|
USHORT *table;
|
|
unsigned int count;
|
|
} rb_code_page;
|
|
|
|
#define IS_DIR_SEPARATOR_P(c) (c == L'\\' || c == L'/')
|
|
#define IS_DIR_UNC_P(c) (IS_DIR_SEPARATOR_P(c[0]) && IS_DIR_SEPARATOR_P(c[1]))
|
|
static int
|
|
IS_ABSOLUTE_PATH_P(const WCHAR *path, size_t len)
|
|
{
|
|
if (len < 2) return FALSE;
|
|
if (ISALPHA(path[0]))
|
|
return len > 2 && path[1] == L':' && IS_DIR_SEPARATOR_P(path[2]);
|
|
else
|
|
return IS_DIR_UNC_P(path);
|
|
}
|
|
|
|
/* MultiByteToWideChar() doesn't work with code page 51932 */
|
|
#define INVALID_CODE_PAGE 51932
|
|
#define PATH_BUFFER_SIZE MAX_PATH * 2
|
|
|
|
#define insecure_obj_p(obj, level) ((level) > 0 && OBJ_TAINTED(obj))
|
|
|
|
/* defined in win32/win32.c */
|
|
#define system_code_page rb_w32_filecp
|
|
#define mbstr_to_wstr rb_w32_mbstr_to_wstr
|
|
#define wstr_to_mbstr rb_w32_wstr_to_mbstr
|
|
|
|
static inline void
|
|
replace_wchar(wchar_t *s, int find, int replace)
|
|
{
|
|
while (*s != 0) {
|
|
if (*s == find)
|
|
*s = replace;
|
|
s++;
|
|
}
|
|
}
|
|
|
|
/* Remove trailing invalid ':$DATA' of the path. */
|
|
static inline size_t
|
|
remove_invalid_alternative_data(wchar_t *wfullpath, size_t size)
|
|
{
|
|
static const wchar_t prime[] = L":$DATA";
|
|
enum { prime_len = (sizeof(prime) / sizeof(wchar_t)) -1 };
|
|
|
|
if (size <= prime_len || _wcsnicmp(wfullpath + size - prime_len, prime, prime_len) != 0)
|
|
return size;
|
|
|
|
/* alias of stream */
|
|
/* get rid of a bug of x64 VC++ */
|
|
if (wfullpath[size - (prime_len + 1)] == ':') {
|
|
/* remove trailing '::$DATA' */
|
|
size -= prime_len + 1; /* prime */
|
|
wfullpath[size] = L'\0';
|
|
}
|
|
else {
|
|
/* remove trailing ':$DATA' of paths like '/aa:a:$DATA' */
|
|
wchar_t *pos = wfullpath + size - (prime_len + 1);
|
|
while (!IS_DIR_SEPARATOR_P(*pos) && pos != wfullpath) {
|
|
if (*pos == L':') {
|
|
size -= prime_len; /* alternative */
|
|
wfullpath[size] = L'\0';
|
|
break;
|
|
}
|
|
pos--;
|
|
}
|
|
}
|
|
return size;
|
|
}
|
|
|
|
void rb_enc_foreach_name(int (*func)(st_data_t name, st_data_t idx, st_data_t arg), st_data_t arg);
|
|
|
|
static int
|
|
code_page_i(st_data_t name, st_data_t idx, st_data_t arg)
|
|
{
|
|
const char *n = (const char *)name;
|
|
if (strncmp("CP", n, 2) == 0) {
|
|
int code_page = atoi(n + 2);
|
|
if (code_page != 0) {
|
|
struct code_page_table *cp = (struct code_page_table *)arg;
|
|
unsigned int count = cp->count;
|
|
USHORT *table = cp->table;
|
|
if (count <= idx) {
|
|
unsigned int i = count;
|
|
count = (((idx + 4) & ~31) | 28);
|
|
table = realloc(table, count * sizeof(*table));
|
|
if (!table) return ST_CONTINUE;
|
|
cp->count = count;
|
|
cp->table = table;
|
|
while (i < count) table[i++] = INVALID_CODE_PAGE;
|
|
}
|
|
table[idx] = (USHORT)code_page;
|
|
}
|
|
}
|
|
return ST_CONTINUE;
|
|
}
|
|
|
|
/*
|
|
Return code page number of the encoding.
|
|
Cache code page into a hash for performance since finding the code page in
|
|
Encoding#names is slow.
|
|
*/
|
|
static UINT
|
|
code_page(rb_encoding *enc)
|
|
{
|
|
int enc_idx;
|
|
|
|
if (!enc)
|
|
return system_code_page();
|
|
|
|
enc_idx = rb_enc_to_index(enc);
|
|
|
|
/* map US-ASCII and ASCII-8bit as code page 1252 (us-ascii) */
|
|
if (enc_idx == rb_usascii_encindex() || enc_idx == rb_ascii8bit_encindex()) {
|
|
return 1252;
|
|
}
|
|
if (enc_idx == rb_utf8_encindex()) {
|
|
return CP_UTF8;
|
|
}
|
|
|
|
if (0 <= enc_idx && (unsigned int)enc_idx < rb_code_page.count)
|
|
return rb_code_page.table[enc_idx];
|
|
|
|
return INVALID_CODE_PAGE;
|
|
}
|
|
|
|
#define fix_string_encoding(str, encoding) rb_str_conv_enc((str), (encoding), rb_utf8_encoding())
|
|
|
|
/*
|
|
Replace the last part of the path to long name.
|
|
We try to avoid to call FindFirstFileW() since it takes long time.
|
|
*/
|
|
static inline size_t
|
|
replace_to_long_name(wchar_t **wfullpath, size_t size, size_t buffer_size)
|
|
{
|
|
WIN32_FIND_DATAW find_data;
|
|
HANDLE find_handle;
|
|
|
|
/*
|
|
Skip long name conversion if the path is already long name.
|
|
Short name is 8.3 format.
|
|
http://en.wikipedia.org/wiki/8.3_filename
|
|
This check can be skipped for directory components that have file
|
|
extensions longer than 3 characters, or total lengths longer than
|
|
12 characters.
|
|
http://msdn.microsoft.com/en-us/library/windows/desktop/aa364980(v=vs.85).aspx
|
|
*/
|
|
size_t const max_short_name_size = 8 + 1 + 3;
|
|
size_t const max_extension_size = 3;
|
|
size_t path_len = 1, extension_len = 0;
|
|
wchar_t *pos = *wfullpath;
|
|
|
|
if (size == 3 && pos[1] == L':' && pos[2] == L'\\' && pos[3] == L'\0') {
|
|
/* root path doesn't need short name expansion */
|
|
return size;
|
|
}
|
|
|
|
/* skip long name conversion if path contains wildcard characters */
|
|
if (wcspbrk(pos, L"*?")) {
|
|
return size;
|
|
}
|
|
|
|
pos = *wfullpath + size - 1;
|
|
while (!IS_DIR_SEPARATOR_P(*pos) && pos != *wfullpath) {
|
|
if (!extension_len && *pos == L'.') {
|
|
extension_len = path_len - 1;
|
|
}
|
|
if (path_len > max_short_name_size || extension_len > max_extension_size) {
|
|
return size;
|
|
}
|
|
path_len++;
|
|
pos--;
|
|
}
|
|
|
|
if ((pos >= *wfullpath + 2) &&
|
|
(*wfullpath)[0] == L'\\' && (*wfullpath)[1] == L'\\') {
|
|
/* UNC path: no short file name, and needs Network Share
|
|
* Management functions instead of FindFirstFile. */
|
|
if (pos == *wfullpath + 2) {
|
|
/* //host only */
|
|
return size;
|
|
}
|
|
if (!wmemchr(*wfullpath + 2, L'\\', pos - *wfullpath - 2)) {
|
|
/* //host/share only */
|
|
return size;
|
|
}
|
|
}
|
|
|
|
find_handle = FindFirstFileW(*wfullpath, &find_data);
|
|
if (find_handle != INVALID_HANDLE_VALUE) {
|
|
size_t trail_pos = pos - *wfullpath + IS_DIR_SEPARATOR_P(*pos);
|
|
size_t file_len = wcslen(find_data.cFileName);
|
|
size_t oldsize = size;
|
|
|
|
FindClose(find_handle);
|
|
size = trail_pos + file_len;
|
|
if (size > (buffer_size ? buffer_size-1 : oldsize)) {
|
|
wchar_t *buf = ALLOC_N(wchar_t, (size + 1));
|
|
wcsncpy(buf, *wfullpath, trail_pos);
|
|
if (!buffer_size)
|
|
xfree(*wfullpath);
|
|
*wfullpath = buf;
|
|
}
|
|
wcsncpy(*wfullpath + trail_pos, find_data.cFileName, file_len + 1);
|
|
}
|
|
return size;
|
|
}
|
|
|
|
static inline size_t
|
|
user_length_in_path(const wchar_t *wuser, size_t len)
|
|
{
|
|
size_t i;
|
|
|
|
for (i = 0; i < len && !IS_DIR_SEPARATOR_P(wuser[i]); i++)
|
|
;
|
|
|
|
return i;
|
|
}
|
|
|
|
static VALUE
|
|
append_wstr(VALUE dst, const WCHAR *ws, ssize_t len, UINT cp, rb_encoding *enc)
|
|
{
|
|
long olen, nlen = (long)len;
|
|
|
|
if (cp != INVALID_CODE_PAGE) {
|
|
if (len == -1) len = lstrlenW(ws);
|
|
nlen = WideCharToMultiByte(cp, 0, ws, len, NULL, 0, NULL, NULL);
|
|
olen = RSTRING_LEN(dst);
|
|
rb_str_modify_expand(dst, nlen);
|
|
WideCharToMultiByte(cp, 0, ws, len, RSTRING_PTR(dst) + olen, nlen, NULL, NULL);
|
|
rb_enc_associate(dst, enc);
|
|
rb_str_set_len(dst, olen + nlen);
|
|
}
|
|
else {
|
|
const int replaceflags = ECONV_UNDEF_REPLACE|ECONV_INVALID_REPLACE;
|
|
char *utf8str = wstr_to_mbstr(CP_UTF8, ws, (int)len, &nlen);
|
|
rb_econv_t *ec = rb_econv_open("UTF-8", rb_enc_name(enc), replaceflags);
|
|
dst = rb_econv_append(ec, utf8str, nlen, dst, replaceflags);
|
|
rb_econv_close(ec);
|
|
free(utf8str);
|
|
}
|
|
return dst;
|
|
}
|
|
|
|
VALUE
|
|
rb_default_home_dir(VALUE result)
|
|
{
|
|
WCHAR *dir = rb_w32_home_dir();
|
|
if (!dir) {
|
|
rb_raise(rb_eArgError, "couldn't find HOME environment -- expanding `~'");
|
|
}
|
|
append_wstr(result, dir, -1,
|
|
rb_w32_filecp(), rb_filesystem_encoding());
|
|
xfree(dir);
|
|
return result;
|
|
}
|
|
|
|
VALUE
|
|
rb_file_expand_path_internal(VALUE fname, VALUE dname, int abs_mode, int long_name, VALUE result)
|
|
{
|
|
size_t size = 0, whome_len = 0;
|
|
size_t buffer_len = 0;
|
|
long wpath_len = 0, wdir_len = 0;
|
|
wchar_t *wfullpath = NULL, *wpath = NULL, *wpath_pos = NULL;
|
|
wchar_t *wdir = NULL, *wdir_pos = NULL;
|
|
wchar_t *whome = NULL, *buffer = NULL, *buffer_pos = NULL;
|
|
UINT path_cp, cp;
|
|
VALUE path = fname, dir = dname;
|
|
wchar_t wfullpath_buffer[PATH_BUFFER_SIZE];
|
|
wchar_t path_drive = L'\0', dir_drive = L'\0';
|
|
int ignore_dir = 0;
|
|
rb_encoding *path_encoding;
|
|
int tainted = 0;
|
|
|
|
/* tainted if path is tainted */
|
|
tainted = OBJ_TAINTED(path);
|
|
|
|
/* get path encoding */
|
|
if (NIL_P(dir)) {
|
|
path_encoding = rb_enc_get(path);
|
|
}
|
|
else {
|
|
path_encoding = rb_enc_check(path, dir);
|
|
}
|
|
|
|
cp = path_cp = code_page(path_encoding);
|
|
|
|
/* workaround invalid codepage */
|
|
if (path_cp == INVALID_CODE_PAGE) {
|
|
cp = CP_UTF8;
|
|
if (!NIL_P(path)) {
|
|
path = fix_string_encoding(path, path_encoding);
|
|
}
|
|
}
|
|
|
|
/* convert char * to wchar_t */
|
|
if (!NIL_P(path)) {
|
|
const long path_len = RSTRING_LEN(path);
|
|
#if SIZEOF_INT < SIZEOF_LONG
|
|
if ((long)(int)path_len != path_len) {
|
|
rb_raise(rb_eRangeError, "path (%ld bytes) is too long",
|
|
path_len);
|
|
}
|
|
#endif
|
|
wpath = mbstr_to_wstr(cp, RSTRING_PTR(path), path_len, &wpath_len);
|
|
wpath_pos = wpath;
|
|
}
|
|
|
|
/* determine if we need the user's home directory */
|
|
/* expand '~' only if NOT rb_file_absolute_path() where `abs_mode` is 1 */
|
|
if (abs_mode == 0 && wpath_len > 0 && wpath_pos[0] == L'~' &&
|
|
(wpath_len == 1 || IS_DIR_SEPARATOR_P(wpath_pos[1]))) {
|
|
/* tainted if expanding '~' */
|
|
tainted = 1;
|
|
|
|
whome = rb_w32_home_dir();
|
|
if (whome == NULL) {
|
|
free(wpath);
|
|
rb_raise(rb_eArgError, "couldn't find HOME environment -- expanding `~'");
|
|
}
|
|
whome_len = wcslen(whome);
|
|
|
|
if (!IS_ABSOLUTE_PATH_P(whome, whome_len)) {
|
|
free(wpath);
|
|
xfree(whome);
|
|
rb_raise(rb_eArgError, "non-absolute home");
|
|
}
|
|
|
|
if (path_cp == INVALID_CODE_PAGE || rb_enc_str_asciionly_p(path)) {
|
|
/* use filesystem encoding if expanding home dir */
|
|
path_encoding = rb_filesystem_encoding();
|
|
cp = path_cp = system_code_page();
|
|
}
|
|
|
|
/* ignores dir since we are expanding home */
|
|
ignore_dir = 1;
|
|
|
|
/* exclude ~ from the result */
|
|
wpath_pos++;
|
|
wpath_len--;
|
|
|
|
/* exclude separator if present */
|
|
if (wpath_len && IS_DIR_SEPARATOR_P(wpath_pos[0])) {
|
|
wpath_pos++;
|
|
wpath_len--;
|
|
}
|
|
}
|
|
else if (wpath_len >= 2 && wpath_pos[1] == L':') {
|
|
if (wpath_len >= 3 && IS_DIR_SEPARATOR_P(wpath_pos[2])) {
|
|
/* ignore dir since path contains a drive letter and a root slash */
|
|
ignore_dir = 1;
|
|
}
|
|
else {
|
|
/* determine if we ignore dir or not later */
|
|
path_drive = wpath_pos[0];
|
|
wpath_pos += 2;
|
|
wpath_len -= 2;
|
|
}
|
|
}
|
|
else if (abs_mode == 0 && wpath_len >= 2 && wpath_pos[0] == L'~') {
|
|
result = rb_str_new_cstr("can't find user ");
|
|
result = append_wstr(result, wpath_pos + 1, user_length_in_path(wpath_pos + 1, wpath_len - 1),
|
|
path_cp, path_encoding);
|
|
|
|
if (wpath)
|
|
free(wpath);
|
|
|
|
rb_exc_raise(rb_exc_new_str(rb_eArgError, result));
|
|
}
|
|
|
|
/* convert dir */
|
|
if (!ignore_dir && !NIL_P(dir)) {
|
|
/* fix string encoding */
|
|
if (path_cp == INVALID_CODE_PAGE) {
|
|
dir = fix_string_encoding(dir, path_encoding);
|
|
}
|
|
|
|
/* convert char * to wchar_t */
|
|
if (!NIL_P(dir)) {
|
|
const long dir_len = RSTRING_LEN(dir);
|
|
#if SIZEOF_INT < SIZEOF_LONG
|
|
if ((long)(int)dir_len != dir_len) {
|
|
if (wpath) free(wpath);
|
|
rb_raise(rb_eRangeError, "base directory (%ld bytes) is too long",
|
|
dir_len);
|
|
}
|
|
#endif
|
|
wdir = mbstr_to_wstr(cp, RSTRING_PTR(dir), dir_len, &wdir_len);
|
|
wdir_pos = wdir;
|
|
}
|
|
|
|
if (abs_mode == 0 && wdir_len > 0 && wdir_pos[0] == L'~' &&
|
|
(wdir_len == 1 || IS_DIR_SEPARATOR_P(wdir_pos[1]))) {
|
|
/* tainted if expanding '~' */
|
|
tainted = 1;
|
|
|
|
whome = rb_w32_home_dir();
|
|
if (whome == NULL) {
|
|
free(wpath);
|
|
free(wdir);
|
|
rb_raise(rb_eArgError, "couldn't find HOME environment -- expanding `~'");
|
|
}
|
|
whome_len = wcslen(whome);
|
|
|
|
if (!IS_ABSOLUTE_PATH_P(whome, whome_len)) {
|
|
free(wpath);
|
|
free(wdir);
|
|
xfree(whome);
|
|
rb_raise(rb_eArgError, "non-absolute home");
|
|
}
|
|
|
|
/* exclude ~ from the result */
|
|
wdir_pos++;
|
|
wdir_len--;
|
|
|
|
/* exclude separator if present */
|
|
if (wdir_len && IS_DIR_SEPARATOR_P(wdir_pos[0])) {
|
|
wdir_pos++;
|
|
wdir_len--;
|
|
}
|
|
}
|
|
else if (wdir_len >= 2 && wdir[1] == L':') {
|
|
dir_drive = wdir[0];
|
|
if (wpath_len && IS_DIR_SEPARATOR_P(wpath_pos[0])) {
|
|
wdir_len = 2;
|
|
}
|
|
}
|
|
else if (wdir_len >= 2 && IS_DIR_UNC_P(wdir)) {
|
|
/* UNC path */
|
|
if (wpath_len && IS_DIR_SEPARATOR_P(wpath_pos[0])) {
|
|
/* cut the UNC path tail to '//host/share' */
|
|
long separators = 0;
|
|
long pos = 2;
|
|
while (pos < wdir_len && separators < 2) {
|
|
if (IS_DIR_SEPARATOR_P(wdir[pos])) {
|
|
separators++;
|
|
}
|
|
pos++;
|
|
}
|
|
if (separators == 2)
|
|
wdir_len = pos - 1;
|
|
}
|
|
}
|
|
else if (abs_mode == 0 && wdir_len >= 2 && wdir_pos[0] == L'~') {
|
|
result = rb_str_new_cstr("can't find user ");
|
|
result = append_wstr(result, wdir_pos + 1, user_length_in_path(wdir_pos + 1, wdir_len - 1),
|
|
path_cp, path_encoding);
|
|
if (wpath)
|
|
free(wpath);
|
|
|
|
if (wdir)
|
|
free(wdir);
|
|
|
|
rb_exc_raise(rb_exc_new_str(rb_eArgError, result));
|
|
}
|
|
}
|
|
|
|
/* determine if we ignore dir or not */
|
|
if (!ignore_dir && path_drive && dir_drive) {
|
|
if (towupper(path_drive) != towupper(dir_drive)) {
|
|
/* ignore dir since path drive is different from dir drive */
|
|
ignore_dir = 1;
|
|
wdir_len = 0;
|
|
dir_drive = 0;
|
|
}
|
|
}
|
|
|
|
if (!ignore_dir && wpath_len >= 2 && IS_DIR_UNC_P(wpath)) {
|
|
/* ignore dir since path has UNC root */
|
|
ignore_dir = 1;
|
|
wdir_len = 0;
|
|
}
|
|
else if (!ignore_dir && wpath_len >= 1 && IS_DIR_SEPARATOR_P(wpath[0]) &&
|
|
!dir_drive && !(wdir_len >= 2 && IS_DIR_UNC_P(wdir))) {
|
|
/* ignore dir since path has root slash and dir doesn't have drive or UNC root */
|
|
ignore_dir = 1;
|
|
wdir_len = 0;
|
|
}
|
|
|
|
buffer_len = wpath_len + 1 + wdir_len + 1 + whome_len + 1;
|
|
|
|
buffer = buffer_pos = ALLOC_N(wchar_t, (buffer_len + 1));
|
|
|
|
/* add home */
|
|
if (whome_len) {
|
|
wcsncpy(buffer_pos, whome, whome_len);
|
|
buffer_pos += whome_len;
|
|
}
|
|
|
|
/* Add separator if required */
|
|
if (whome_len && wcsrchr(L"\\/:", buffer_pos[-1]) == NULL) {
|
|
buffer_pos[0] = L'\\';
|
|
buffer_pos++;
|
|
}
|
|
else if (!dir_drive && path_drive) {
|
|
*buffer_pos++ = path_drive;
|
|
*buffer_pos++ = L':';
|
|
}
|
|
|
|
if (wdir_len) {
|
|
/* tainted if dir is used and dir is tainted */
|
|
if (!tainted && OBJ_TAINTED(dir))
|
|
tainted = 1;
|
|
|
|
wcsncpy(buffer_pos, wdir_pos, wdir_len);
|
|
buffer_pos += wdir_len;
|
|
}
|
|
|
|
/* add separator if required */
|
|
if (wdir_len && wcsrchr(L"\\/:", buffer_pos[-1]) == NULL) {
|
|
buffer_pos[0] = L'\\';
|
|
buffer_pos++;
|
|
}
|
|
|
|
/* now deal with path */
|
|
if (wpath_len) {
|
|
wcsncpy(buffer_pos, wpath_pos, wpath_len);
|
|
buffer_pos += wpath_len;
|
|
}
|
|
|
|
/* GetFullPathNameW requires at least "." to determine current directory */
|
|
if (wpath_len == 0) {
|
|
buffer_pos[0] = L'.';
|
|
buffer_pos++;
|
|
}
|
|
|
|
/* Ensure buffer is NULL terminated */
|
|
buffer_pos[0] = L'\0';
|
|
|
|
/* tainted if path is relative */
|
|
if (!tainted && !IS_ABSOLUTE_PATH_P(buffer, buffer_len))
|
|
tainted = 1;
|
|
|
|
/* FIXME: Make this more robust */
|
|
/* Determine require buffer size */
|
|
size = GetFullPathNameW(buffer, PATH_BUFFER_SIZE, wfullpath_buffer, NULL);
|
|
if (size > PATH_BUFFER_SIZE) {
|
|
/* allocate more memory than allotted originally by PATH_BUFFER_SIZE */
|
|
wfullpath = ALLOC_N(wchar_t, size);
|
|
size = GetFullPathNameW(buffer, size, wfullpath, NULL);
|
|
}
|
|
else {
|
|
wfullpath = wfullpath_buffer;
|
|
}
|
|
|
|
/* Remove any trailing slashes */
|
|
if (IS_DIR_SEPARATOR_P(wfullpath[size - 1]) &&
|
|
wfullpath[size - 2] != L':' &&
|
|
!(size == 2 && IS_DIR_UNC_P(wfullpath))) {
|
|
size -= 1;
|
|
wfullpath[size] = L'\0';
|
|
}
|
|
|
|
/* Remove any trailing dot */
|
|
if (wfullpath[size - 1] == L'.') {
|
|
size -= 1;
|
|
wfullpath[size] = L'\0';
|
|
}
|
|
|
|
/* removes trailing invalid ':$DATA' */
|
|
size = remove_invalid_alternative_data(wfullpath, size);
|
|
|
|
/* Replace the trailing path to long name */
|
|
if (long_name) {
|
|
size_t bufsize = wfullpath == wfullpath_buffer ? PATH_BUFFER_SIZE : 0;
|
|
size = replace_to_long_name(&wfullpath, size, bufsize);
|
|
}
|
|
|
|
/* sanitize backslashes with forwardslashes */
|
|
replace_wchar(wfullpath, L'\\', L'/');
|
|
|
|
/* convert to VALUE and set the path encoding */
|
|
rb_str_set_len(result, 0);
|
|
result = append_wstr(result, wfullpath, size, path_cp, path_encoding);
|
|
|
|
/* makes the result object tainted if expanding tainted strings or returning modified path */
|
|
if (tainted)
|
|
OBJ_TAINT(result);
|
|
|
|
/* TODO: better cleanup */
|
|
if (buffer)
|
|
xfree(buffer);
|
|
|
|
if (wpath)
|
|
free(wpath);
|
|
|
|
if (wdir)
|
|
free(wdir);
|
|
|
|
if (whome)
|
|
xfree(whome);
|
|
|
|
if (wfullpath != wfullpath_buffer)
|
|
xfree(wfullpath);
|
|
|
|
rb_enc_associate(result, path_encoding);
|
|
return result;
|
|
}
|
|
|
|
VALUE
|
|
rb_readlink(VALUE path, rb_encoding *resultenc)
|
|
{
|
|
DWORD len;
|
|
VALUE wtmp = 0, wpathbuf, str;
|
|
rb_w32_reparse_buffer_t rbuf, *rp = &rbuf;
|
|
WCHAR *wpath, *wbuf;
|
|
rb_encoding *enc;
|
|
UINT cp, path_cp;
|
|
int e;
|
|
|
|
FilePathValue(path);
|
|
enc = rb_enc_get(path);
|
|
cp = path_cp = code_page(enc);
|
|
if (cp == INVALID_CODE_PAGE) {
|
|
path = fix_string_encoding(path, enc);
|
|
cp = CP_UTF8;
|
|
}
|
|
len = MultiByteToWideChar(cp, 0, RSTRING_PTR(path), RSTRING_LEN(path), NULL, 0);
|
|
wpath = ALLOCV_N(WCHAR, wpathbuf, len+1);
|
|
MultiByteToWideChar(cp, 0, RSTRING_PTR(path), RSTRING_LEN(path), wpath, len);
|
|
wpath[len] = L'\0';
|
|
e = rb_w32_read_reparse_point(wpath, rp, sizeof(rbuf), &wbuf, &len);
|
|
if (e == ERROR_MORE_DATA) {
|
|
size_t size = rb_w32_reparse_buffer_size(len + 1);
|
|
rp = ALLOCV(wtmp, size);
|
|
e = rb_w32_read_reparse_point(wpath, rp, size, &wbuf, &len);
|
|
}
|
|
ALLOCV_END(wpathbuf);
|
|
if (e) {
|
|
ALLOCV_END(wtmp);
|
|
if (e != -1)
|
|
rb_syserr_fail_path(rb_w32_map_errno(e), path);
|
|
else /* not symlink; maybe volume mount point */
|
|
rb_syserr_fail_path(EINVAL, path);
|
|
}
|
|
enc = resultenc;
|
|
path_cp = code_page(enc);
|
|
len = lstrlenW(wbuf);
|
|
str = append_wstr(rb_enc_str_new(0, 0, enc), wbuf, len, path_cp, enc);
|
|
ALLOCV_END(wtmp);
|
|
return str;
|
|
}
|
|
|
|
int
|
|
rb_file_load_ok(const char *path)
|
|
{
|
|
DWORD attr;
|
|
int ret = 1;
|
|
long len;
|
|
wchar_t* wpath;
|
|
|
|
wpath = mbstr_to_wstr(CP_UTF8, path, -1, &len);
|
|
if (!wpath) return 0;
|
|
|
|
attr = GetFileAttributesW(wpath);
|
|
if (attr == INVALID_FILE_ATTRIBUTES ||
|
|
(attr & FILE_ATTRIBUTE_DIRECTORY)) {
|
|
ret = 0;
|
|
}
|
|
else {
|
|
HANDLE h = CreateFileW(wpath, GENERIC_READ,
|
|
FILE_SHARE_READ | FILE_SHARE_WRITE,
|
|
NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);
|
|
if (h != INVALID_HANDLE_VALUE) {
|
|
CloseHandle(h);
|
|
}
|
|
else {
|
|
ret = 0;
|
|
}
|
|
}
|
|
free(wpath);
|
|
return ret;
|
|
}
|
|
|
|
int
|
|
rb_freopen(VALUE fname, const char *mode, FILE *file)
|
|
{
|
|
WCHAR *wname, wmode[4];
|
|
VALUE wtmp;
|
|
char *name;
|
|
long len;
|
|
int e = 0, n = MultiByteToWideChar(CP_ACP, 0, mode, -1, NULL, 0);
|
|
if (n > numberof(wmode)) return EINVAL;
|
|
MultiByteToWideChar(CP_ACP, 0, mode, -1, wmode, numberof(wmode));
|
|
RSTRING_GETMEM(fname, name, len);
|
|
n = rb_long2int(len);
|
|
len = MultiByteToWideChar(CP_UTF8, 0, name, n, NULL, 0);
|
|
wname = ALLOCV_N(WCHAR, wtmp, len + 1);
|
|
len = MultiByteToWideChar(CP_UTF8, 0, name, n, wname, len);
|
|
wname[len] = L'\0';
|
|
RB_GC_GUARD(fname);
|
|
#if RUBY_MSVCRT_VERSION < 80 && !defined(HAVE__WFREOPEN_S)
|
|
e = _wfreopen(wname, wmode, file) ? 0 : errno;
|
|
#else
|
|
{
|
|
FILE *newfp = 0;
|
|
e = _wfreopen_s(&newfp, wname, wmode, file);
|
|
}
|
|
#endif
|
|
ALLOCV_END(wtmp);
|
|
return e;
|
|
}
|
|
|
|
void
|
|
Init_w32_codepage(void)
|
|
{
|
|
if (rb_code_page.count) return;
|
|
rb_enc_foreach_name(code_page_i, (st_data_t)&rb_code_page);
|
|
}
|