[PATCH] Expand the rename(2) workaround in mingw to cover case change in filename

Windows, being a case-confused operating system, sometimes has
problems renaming directory entries with only change in the case
of their characters. Try to work the problem around by using an
intermediate file.

---

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

And sorry for somewhat extended Cc:. Thought the problem need a little
more experienced attention (then again, maybe it doesn't).

That's a response to Bug 228 (which is properly "WONTFIX",
if you ask my opinion):

    http://code.google.com/p/msysgit/issues/detail?id=228

I don't like it, but it is the same what people usually forced to do
anyway, so here it is.

It has a racing problem which can cause the original file be renamed
into a temporary name (as returned by mkstemp):

    - the rename of the source name to temp name succeeds
    - the rename of temp name to destination name fails (i.e. it
      exists and is open)

It is not fixable (because the atomicity of rename(2) is broken by the
intermediate name and two separate calls to rename involved).
I could have tried to rename back to source, but got fed up with
another Windows stupidity and gave up.

 compat/mingw.c |   65 +++++++++++++++++++++++++++++++++++++++++++++++--------
 1 files changed, 55 insertions(+), 10 deletions(-)

diff --git a/compat/mingw.c b/compat/mingw.c
index e190fdd..ad60dff 100644
--- a/compat/mingw.c
+++ b/compat/mingw.c

@@ -931,21 +931,12 @@ int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz)
 	return connect(s, sa, sz);
 }
 
-#undef rename
-int mingw_rename(const char *pold, const char *pnew)
+static int move_file_replace(const char *pold, const char *pnew)
 {
 	DWORD attrs, gle;
 	int tries = 0;
 	static const int delay[] = { 0, 1, 10, 20, 40 };
 
-	/*
-	 * Try native rename() first to get errno right.
-	 * It is based on MoveFile(), which cannot overwrite existing files.
-	 */
-	if (!rename(pold, pnew))
-		return 0;
-	if (errno != EEXIST)
-		return -1;
 repeat:
 	if (MoveFileEx(pold, pnew, MOVEFILE_REPLACE_EXISTING))
 		return 0;

@@ -982,6 +973,60 @@ repeat:
 	return -1;
 }
 
+#undef rename
+int mingw_rename(const char *pold, const char *pnew)
+{
+	int fd, rc = -1;
+	char *p, *tmp = NULL;
+
+	/*
+	 * Try native rename() first to get errno right.
+	 * It is based on MoveFile(), which cannot overwrite existing files.
+	 */
+	if (!rename(pold, pnew))
+		return 0;
+	if (errno != EEXIST)
+		return -1;
+	/*
+	 * Windows' case-insensitivity does not allow it to directly
+	 * do a rename where the only change in the file name is
+	 * the change of a letter case. Work this around with a
+	 * temporary file.
+	 */
+	if (!dst) {
+		errno = EINVAL;
+		goto fail;
+	}
+	tmp = malloc(strlen(dst) + 7 /* reserved for template */);
+	if (!tmp) {
+		errno = ENOMEM;
+		goto fail;
+	}
+	strcpy(tmp, dst);
+	p = tmp + strlen(dst);
+	for (p = tmp + strlen(dst); p > tmp; --p)
+		if ('\\' == *p || '/' == *p) {
+			++p;
+			break;
+		}
+	strcpy(p, "XXXXXX")
+	fd = mkstemp(tmp);
+	if (fd < 0)
+		goto fail;
+	close(fd);
+	if (move_file_replace(src, tmp)) {
+		rc = errno;
+		unlink(tmp);
+		errno = rc;
+		rc = -1;
+		goto fail;
+	}
+	rc = move_file_replace(tmp, dst);
+fail:
+	free(tmp);
+	return rc;
+}
+
 struct passwd *getpwuid(int uid)
 {
 	static char user_name[100];

-- 
1.6.3.2.214.g82a9d

Hi,

On Mon, 8 Jun 2009, Alex Riesen wrote:

Windows, being a case-confused operating system, sometimes has
problems renaming directory entries with only change in the case
of their characters. Try to work the problem around by using an
intermediate file.

---

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then 
src/dst.

I minimally fixed up your patch (it now uses strbuf), and added a 
test-case.

This test-case is actually crucial: it shows that your patch is not 
enough: the culprit is this code in builtin-mv.c:167--168:

                else if (lstat(dst, &st) == 0) {
                        bad = "destination exists";

Ciao,
Dscho

Hi,

On Mon, 8 Jun 2009, Johannes Schindelin wrote:

On Mon, 8 Jun 2009, Alex Riesen wrote:

quoted

Windows, being a case-confused operating system, sometimes has
problems renaming directory entries with only change in the case
of their characters. Try to work the problem around by using an
intermediate file.

---

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then 
src/dst.

I minimally fixed up your patch (it now uses strbuf), and added a 
test-case.

I probably should have mentioned that you can see it here:

http://repo.or.cz/w/git/mingw/4msysgit.git?a=shortlog;h=refs/heads/work/rename

Ciao,
Dscho

2009/6/8 Johannes Schindelin [off-list ref]:

On Mon, 8 Jun 2009, Alex Riesen wrote:

quoted

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then
src/dst.

Yep.

I minimally fixed up your patch (it now uses strbuf), and added a
test-case.

This test-case is actually crucial: it shows that your patch is not
enough: the culprit is this code in builtin-mv.c:167--168:

               else if (lstat(dst, &st) == 0) {
                       bad = "destination exists";

Ah, thanks. Missed that completely.

Alex Riesen schrieb:

2009/6/8 Johannes Schindelin [off-list ref]:

quoted

On Mon, 8 Jun 2009, Alex Riesen wrote:

quoted

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then
src/dst.

Yep.

quoted

I minimally fixed up your patch (it now uses strbuf), and added a
test-case.

This test-case is actually crucial: it shows that your patch is not
enough: the culprit is this code in builtin-mv.c:167--168:

               else if (lstat(dst, &st) == 0) {
                       bad = "destination exists";

Ah, thanks. Missed that completely.

That's the reason why I think any work-around for this problem is not
worth it.

If you want to be cross-platfrom, make up your mind about file names in
advance.

-- Hannes

2009/6/9 Johannes Sixt [off-list ref]:

Alex Riesen schrieb:

quoted

2009/6/8 Johannes Schindelin [off-list ref]:

quoted

On Mon, 8 Jun 2009, Alex Riesen wrote:

quoted

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then
src/dst.

Yep.

quoted

I minimally fixed up your patch (it now uses strbuf), and added a
test-case.

This test-case is actually crucial: it shows that your patch is not
enough: the culprit is this code in builtin-mv.c:167--168:

               else if (lstat(dst, &st) == 0) {
                       bad = "destination exists";

Ah, thanks. Missed that completely.

That's the reason why I think any work-around for this problem is not
worth it.

What is the lstat is for, anyway?

If you want to be cross-platfrom, make up your mind about file names in
advance.

I suspect it is the other way around. Some platforms just hate the idea
of being ported to. Or its designers have poor imagination and never
though about the fact that other platforms exist.

Hi,

On Tue, 9 Jun 2009, Alex Riesen wrote:

2009/6/9 Johannes Sixt [off-list ref]:

quoted

Alex Riesen schrieb:

quoted

2009/6/8 Johannes Schindelin [off-list ref]:

quoted

On Mon, 8 Jun 2009, Alex Riesen wrote:

quoted

NOT TESTED. Can't. My Windows broke again. Not even compile-tested.

Yes, that is pretty easy to see as you first used pold/pnew and then
src/dst.

Yep.

quoted

I minimally fixed up your patch (it now uses strbuf), and added a
test-case.

This test-case is actually crucial: it shows that your patch is not
enough: the culprit is this code in builtin-mv.c:167--168:

               else if (lstat(dst, &st) == 0) {
                       bad = "destination exists";

Ah, thanks. Missed that completely.

That's the reason why I think any work-around for this problem is not
worth it.

What is the lstat is for, anyway?

It is to avoid overwriting existing files.

quoted

If you want to be cross-platfrom, make up your mind about file names 
in advance.

I suspect it is the other way around. Some platforms just hate the idea 
of being ported to. Or its designers have poor imagination and never 
though about the fact that other platforms exist.

Actually, think about it as a brilliant business plan: make developers be 
stuck with developing for your platform, then they lack the time to take 
care of other platforms.

The bigger question with the patch is if the lstat() call can tell us 
if this is the same file (really the same, not just a hardlink (= same 
contents)).

Ciao,
Dscho