git-commit-vandalism/builtin-rm.c

155 lines
3.2 KiB
C
Raw Normal View History

Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
/*
* "git rm" builtin command
*
* Copyright (C) Linus Torvalds 2006
*/
#include "cache.h"
#include "builtin.h"
#include "dir.h"
#include "cache-tree.h"
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
static const char builtin_rm_usage[] =
"git-rm [-n] [-v] [-f] <filepattern>...";
static struct {
int nr, alloc;
const char **name;
} list;
static void add_list(const char *name)
{
if (list.nr >= list.alloc) {
list.alloc = alloc_nr(list.alloc);
list.name = xrealloc(list.name, list.alloc * sizeof(const char *));
}
list.name[list.nr++] = name;
}
static int remove_file(const char *name)
{
int ret;
char *slash;
ret = unlink(name);
if (!ret && (slash = strrchr(name, '/'))) {
char *n = strdup(name);
do {
n[slash - name] = 0;
name = n;
} while (!rmdir(name) && (slash = strrchr(name, '/')));
}
return ret;
}
static struct lock_file lock_file;
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
int cmd_rm(int argc, const char **argv, char **envp)
{
int i, newfd;
int verbose = 0, show_only = 0, force = 0;
const char *prefix = setup_git_directory();
const char **pathspec;
char *seen;
git_config(git_default_config);
newfd = hold_lock_file_for_update(&lock_file, get_index_file());
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
if (newfd < 0)
die("unable to create new index file");
if (read_cache() < 0)
die("index file corrupt");
for (i = 1 ; i < argc ; i++) {
const char *arg = argv[i];
if (*arg != '-')
break;
if (!strcmp(arg, "--")) {
i++;
break;
}
if (!strcmp(arg, "-n")) {
show_only = 1;
continue;
}
if (!strcmp(arg, "-v")) {
verbose = 1;
continue;
}
if (!strcmp(arg, "-f")) {
force = 1;
continue;
}
die(builtin_rm_usage);
}
if (argc <= i)
usage(builtin_rm_usage);
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
pathspec = get_pathspec(prefix, argv + i);
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
seen = NULL;
for (i = 0; pathspec[i] ; i++)
/* nothing */;
seen = xcalloc(i, 1);
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
for (i = 0; i < active_nr; i++) {
struct cache_entry *ce = active_cache[i];
if (!match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))
continue;
add_list(ce->name);
}
if (pathspec) {
const char *match;
for (i = 0; (match = pathspec[i]) != NULL ; i++) {
if (*match && !seen[i])
die("pathspec '%s' did not match any files", match);
}
}
/*
* First remove the names from the index: we won't commit
* the index unless all of them succeed
*/
for (i = 0; i < list.nr; i++) {
const char *path = list.name[i];
printf("rm '%s'\n", path);
if (remove_file_from_cache(path))
die("git rm: unable to remove %s", path);
cache_tree_invalidate_path(active_cache_tree, path);
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
}
if (show_only)
return 0;
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
/*
* Then, if we used "-f", remove the filenames from the
* workspace. If we fail to remove the first one, we
* abort the "git rm" (but once we've successfully removed
* any file at all, we'll go ahead and commit to it all:
* by then we've already committed ourselves and can't fail
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
* in the middle)
*/
if (force) {
int removed = 0;
for (i = 0; i < list.nr; i++) {
const char *path = list.name[i];
if (!remove_file(path)) {
removed = 1;
continue;
}
if (!removed)
die("git rm: %s: %s", path, strerror(errno));
}
}
if (active_cache_changed) {
if (write_cache(newfd, active_cache, active_nr) ||
close(newfd) || commit_lock_file(&lock_file))
Add builtin "git rm" command This changes semantics very subtly, because it adds a new atomicity guarantee. In particular, if you "git rm" several files, it will now do all or nothing. The old shell-script really looped over the removed files one by one, and would basically randomly fail in the middle if "-f" was used and one of the files didn't exist in the working directory. This C builtin one will not re-write the index after each remove, but instead remove all files at once. However, that means that if "-f" is used (to also force removal of the file from the working directory), and some files have already been removed from the workspace, it won't stop in the middle in some half-way state like the old one did. So what happens is that if the _first_ file fails to be removed with "-f", we abort the whole "git rm". But once we've started removing, we don't leave anything half done. If some of the other files don't exist, we'll just ignore errors of removal from the working tree. This is only an issue with "-f", of course. I think the new behaviour is strictly an improvement, but perhaps more importantly, it is _different_. As a special case, the semantics are identical for the single-file case (which is the only one our test-suite seems to test). The other question is what to do with leading directories. The old "git rm" script didn't do anything, which is somewhat inconsistent. This one will actually clean up directories that have become empty as a result of removing the last file, but maybe we want to have a flag to decide the behaviour? Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-05-20 01:19:34 +02:00
die("Unable to write new index file");
}
return 0;
}