X-Git-Url: http://git.vanrenterghem.biz/git.ikiwiki.info.git/blobdiff_plain/c96d672810ae524aedf9ffff08dcca773cd9a876..2bb049df4fe55191783e6625dbac719e52e329fd:/doc/todo/Moving_Pages.mdwn
diff --git a/doc/todo/Moving_Pages.mdwn b/doc/todo/Moving_Pages.mdwn
index 2e0603ca7..387e4fb82 100644
--- a/doc/todo/Moving_Pages.mdwn
+++ b/doc/todo/Moving_Pages.mdwn
@@ -10,6 +10,8 @@ Thanks again to [Joey](http://kitenet.net/~joey) for putting ikiwiki together.
*[Kyle](http://kitenet.net/~kyle/)=*
+> Took a bit too long, but [[done]] now. --[[Joey]]
+
----
The MediaWiki moving/renaming mechanism is pretty nice. It's easy to get a list of pages that point to the current page. When renaming a page it sticks a forwarding page in the original place. The larger the size of the wiki the more important organization tools become.
@@ -30,286 +32,191 @@ Brad
>
> This could also be implemented using a combination of raw inline and meta
> to change the title (add a "redirected from etc." page. This could be done
-> with a plugin. A redirect page would be [[redirect page="newpage"]].
+> with a plugin. A redirect page would be [[!redirect page="newpage"]].
> But then if you click "edit" on this redirect page, you won't be able
> to edit the new page, only the call to redirect.
> --Ethan
-----
-[[tag patch]]
-
-This is my second cut at a feature like that requested here.
-It can also be found [here](http://ikidev.betacantrips.com/patches/move.patch).
-
-A few shortcomings exist:
-
-* No precautions whatsoever are made to protect against race conditions or failures
- in the rcs\_move function. I didn't even do the `cgi_editpage` thing where I hold
- the lock and render afterwards (mostly because the copy I was editing was not
- up-to-date enough to have that code). Although FAILED_SAVE is in movepage.tmpl,
- no code activates it yet.
-* Some code is duplicated between cgi\_movepage and cgi\_editpage, as well
- as rcs\_commit and rcs\_move.
-* The user interface is pretty lame. I couldn't figure out a good way to let
- the user specify which directory to move things to without implementing a
- FileChooser thing.
-* No redirect pages like those mentioned on [[todo/Moving_Pages]] exist yet,
- so none are created.
-* I added a Move link to page.tmpl but it may belong better someplace else --
- maybe editpage.tmpl? Not sure.
-* from is redundant with page so far -- but since the Move links could someday
- come from someplace other than the page itself I kept it around.
-* If I move foo.mdwn to bar.mdwn, foo/* should move too, probably.
-
-> Looks like a good start, although I agree about many of the points above,
-> and also feel that something needs to be done about rcses that don't
-> implement a move operation -- falling back to an add and delete.
-> --[[Joey]]
-
-Hmm. Shouldn't that be done on a by-RCS basis, though? (i.e. implemented
-by backends in the `rcs_move` function)
-
-> Probably, yes, but maybe there's a way to avoid duplicating code for that
-> in several of them.
-
-Also, how should ikiwiki react if a page is edited (say, by another user)
-before it is moved? Bail, or shrug and proceed?
-
-> The important thing is to keep in mind that the page could be edited,
-> moved, deleted, etc in between the user starting the move and the move
-> happening. So, the code really needs to deal with all of these cases in
-> some way. It seems fine to me to go ahead with the move even if the page
-> was edited. If the page was deleted or moved, it seems reasonable to exit
-> with an error.
-
- diff -urNX ignorepats ikiwiki/IkiWiki/CGI.pm ikidev/IkiWiki/CGI.pm
- --- ikiwiki/IkiWiki/CGI.pm 2007-02-14 18:17:12.000000000 -0800
- +++ ikidev/IkiWiki/CGI.pm 2007-02-22 18:54:23.194982000 -0800
- @@ -561,6 +561,106 @@
- }
- } #}}}
-
- +sub cgi_movepage($$) {
- + my $q = shift;
- + my $session = shift;
- + eval q{use CGI::FormBuilder};
- + error($@) if $@;
- + my @fields=qw(do from rcsinfo page newdir newname comments);
- + my @buttons=("Rename Page", "Cancel");
- +
- + my $form = CGI::FormBuilder->new(
- + fields => \@fields,
- + header => 1,
- + charset => "utf-8",
- + method => 'POST',
- + action => $config{cgiurl},
- + template => (-e "$config{templatedir}/movepage.tmpl" ?
- + {template_params("movepage.tmpl")} : ""),
- + );
- + run_hooks(formbuilder_setup => sub {
- + shift->(form => $form, cgi => $q, session => $session);
- + });
- +
- + decode_form_utf8($form);
- +
- + # This untaint is safe because if the page doesn't exist, bail.
- + my $page = $form->field('page');
- + $page = possibly_foolish_untaint($page);
- + if (! exists $pagesources{$page}) {
- + error("page does not exist");
- + }
- + my $file=$pagesources{$page};
- + my $type=pagetype($file);
- +
- + my $from;
- + if (defined $form->field('from')) {
- + ($from)=$form->field('from')=~/$config{wiki_file_regexp}/;
- + }
- +
- + $form->field(name => "do", type => 'hidden');
- + $form->field(name => "from", type => 'hidden');
- + $form->field(name => "rcsinfo", type => 'hidden');
- + $form->field(name => "newdir", type => 'text', size => 80);
- + $form->field(name => "page", value => $page, force => 1);
- + $form->field(name => "newname", type => "text", size => 80);
- + $form->field(name => "comments", type => "text", size => 80);
- + $form->tmpl_param("can_commit", $config{rcs});
- + $form->tmpl_param("indexlink", indexlink());
- + $form->tmpl_param("baseurl", baseurl());
- +
- + if (! $form->submitted) {
- + $form->field(name => "rcsinfo", value => rcs_prepedit($file),
- + force => 1);
- + }
- +
- + if ($form->submitted eq "Cancel") {
- + redirect($q, "$config{url}/".htmlpage($page));
- + return;
- + }
- +
- + if (! $form->submitted || ! $form->validate) {
- + check_canedit($page, $q, $session);
- + $form->tmpl_param("page_select", 0);
- + $form->field(name => "page", type => 'hidden');
- + $form->field(name => "type", type => 'hidden');
- + $form->title(sprintf(gettext("moving %s"), pagetitle($page)));
- + my $pname = basename($page);
- + my $dname = dirname($page);
- + if (! defined $form->field('newname') ||
- + ! length $form->field('newname')) {
- + $form->field(name => "newname",
- + value => pagetitle($pname, 1), force => 1);
- + }
- + if (! defined $form->field('newdir') ||
- + ! length $form->field('newdir')) {
- + $form->field(name => "newdir",
- + value => pagetitle($dname, 1), force => 1);
- + }
- + print $form->render(submit => \@buttons);
- + }
- + else{
- + # This untaint is safe because titlepage removes any problematic
- + # characters.
- + my ($newname)=$form->field('newname');
- + $newname=titlepage(possibly_foolish_untaint($newname));
- + my ($newdir)=$form->field('newdir');
- + $newdir=titlepage(possibly_foolish_untaint($newdir));
- + if (! defined $newname || ! length $newname || file_pruned($newname, $config{srcdir}) || $newname=~/^\//) {
- + error("bad page name");
- + }
- + check_canedit($page, $q, $session);
- +
- + my $newpage = ($newdir?"$newdir/":"") . $newname;
- + my $newfile = $newpage . ".$type";
- + my $message = $form->field('comments');
- + unlockwiki();
- + rcs_move($file, $newfile, $message, $form->field("rcsinfo"),
- + $session->param("name"), $ENV{REMOTE_ADDR});
- + redirect($q, "$config{url}/".htmlpage($newpage));
- + }
- +}
- +
- sub cgi_getsession ($) { #{{{
- my $q=shift;
-
- @@ -656,6 +756,9 @@
- elsif (defined $session->param("postsignin")) {
- cgi_postsignin($q, $session);
- }
- + elsif ($do eq 'move') {
- + cgi_movepage($q, $session);
- + }
- elsif ($do eq 'prefs') {
- cgi_prefs($q, $session);
- }
- diff -urNX ignorepats ikiwiki/IkiWiki/Rcs/svn.pm ikidev/IkiWiki/Rcs/svn.pm
- --- ikiwiki/IkiWiki/Rcs/svn.pm 2007-01-27 16:04:48.000000000 -0800
- +++ ikidev/IkiWiki/Rcs/svn.pm 2007-02-22 01:51:29.923626000 -0800
- @@ -60,6 +60,34 @@
- }
- } #}}}
-
- +sub rcs_move ($$$$;$$) {
- + my $file=shift;
- + my $newname=shift;
- + my $message=shift;
- + my $rcstoken=shift;
- + my $user=shift;
- + my $ipaddr=shift;
- + if (defined $user) {
- + $message="web commit by $user".(length $message ? ": $message" : "");
- + }
- + elsif (defined $ipaddr) {
- + $message="web commit from $ipaddr".(length $message ? ": $message" : "");
- + }
- +
- + chdir($config{srcdir}); # svn merge wants to be here
- +
- + if (system("svn", "move", "--quiet",
- + "$file", "$newname") != 0) {
- + return 1;
- + }
- + if (system("svn", "commit", "--quiet",
- + "--encoding", "UTF-8", "-m",
- + possibly_foolish_untaint($message)) != 0) {
- + return 1;
- + }
- + return undef # success
- +}
- +
- sub rcs_commit ($$$;$$) { #{{{
- # Tries to commit the page; returns undef on _success_ and
- # a version of the page with the rcs's conflict markers on failure.
- diff -urNX ignorepats ikiwiki/IkiWiki/Render.pm ikidev/IkiWiki/Render.pm
- --- ikiwiki/IkiWiki/Render.pm 2007-02-14 17:00:05.000000000 -0800
- +++ ikidev/IkiWiki/Render.pm 2007-02-22 18:30:00.451755000 -0800
- @@ -80,6 +80,7 @@
-
- if (length $config{cgiurl}) {
- $template->param(editurl => cgiurl(do => "edit", page => $page));
- + $template->param(moveurl => cgiurl(do => "move", page => $page));
- $template->param(prefsurl => cgiurl(do => "prefs"));
- if ($config{rcs}) {
- $template->param(recentchangesurl => cgiurl(do => "recentchanges"));
- diff -urNX ignorepats ikiwiki/templates/movepage.tmpl ikidev/templates/movepage.tmpl
- --- ikiwiki/templates/movepage.tmpl 1969-12-31 16:00:00.000000000 -0800
- +++ ikidev/templates/movepage.tmpl 2007-02-22 18:40:39.751763000 -0800
- @@ -0,0 +1,44 @@
- +
- +
- +
- +
- +
- +<TMPL_VAR FORM-TITLE>
- +
- +
- +
- +
- +
- +
- +
- +
- +
- +Failed to save your changes.
- +
- +
- +Your changes were not able to be saved to disk. The system gave the error:
- +
- +
- +
- +Your changes are preserved below, and you can try again to save them.
- +
- +
- +
- +
-
+I'm going to try to run through a full analysis and design for moving and
+deleting pages here. I want to make sure all cases are covered. --[[Joey]]
+
+## UI
+
+The UI I envision is to add "Rename" and "Delete" buttons to the file edit
+page. Both next to the Save button, and also at the bottom of the attachment
+management interface.
+
+The attachment(s) to rename or delete would be selected using the check boxes
+and then the button applies to all of them. Deleting multiple attachments
+in one go is fine; renaming multiple attachments in one go is ambiguous,
+and it can just error out if more than one is selected for rename.
+(Alternatively, it could allow moving them all to a different subdirectory.)
+
+The Delete buttons lead to a page to confirm the deletion(s).
+
+The Rename buttons lead to a page with a text edit box for editing the
+page name. The title of the page is edited, not the actual filename.
+
+There will also be a optional comment field, so a commit message can be
+written for the rename/delete.
+
+Note that there's an edge case concerning pages that have a "/" encoded
+as part of their title. There's no way for a title edit box to
+differentiate between that, and a "/" that is instended to refer to a
+subdirectory to move the page to. Consequence is that "/" will always be
+treated literally, as a subdir separator; it will not be possible to use
+this interface to put an encoded "/" in a page's name.
+
+Once a page is renamed, ikiwiki will return to the page edit interface,
+now for the renamed page. Any modifications that the user had made to the
+textarea will be preserved.
+
+Similarly, when an attachment is renamed, or deleted, return to the page
+edit interface (with the attachments displayed).
+
+When a page is deleted, redirect the user to the toplevel index.
+
+Note that this design, particularly the return to the edit interface after
+rename, means that the rename button can *only* be put on the page edit ui.
+It won't be possible to put it on the action bar or somewhere else. (It
+would be possible to code up a different rename button that doesn't do
+that, and use it elsewhere.)
+
+Hmm, unless it saves the edit state and reloads it later, while using a separate
+form. Which seems to solve other problems, so I think is the way to go.
+
+## SubPages
+
+When renaming `foo`, it probably makes sense to also rename
+`foo/Discussion`. Should other SubPages in `foo/` also be renamed? I think
+it's probably simplest to rename all of its SubPages too.
+
+(For values of "simplest" that don't include the pain of dealing with all
+the changed links on subpages.. as well as issues like pagespecs that
+continue to match the old subpages, and cannot reasonably be auto-converted
+to use the new, etc, etc... So still undecided about this.)
+
+When deleting `foo`, I don't think SubPages should be deleted. The
+potential for mistakes and abuse is too large. Deleting Discussion page
+might be a useful exception.
+
+TODO: Currently, subpages are not addressed.
+
+## link fixups
+
+When renaming a page, it's desirable to keep links that point to it
+working. Rather than use redirection pages, I think that all pages that
+link to it should be modified to fix their links.
+
+The rename plugin can add a "rename" hook, which other plugins can use to
+update links &etc. The hook would be passed page content, the old and new
+link names, and would modify the content and return it. At least the link
+plugin should have such a hook.
+
+After calling the "rename" hook, and rendering the wiki, the rename plugin
+can check to see what links remain pointing to the old page. There could
+still be some, for example, CamelCase links probably won't be changed; img
+plugins and others contain logical links to the file, etc. The user can be
+presented with a list of all the pages that still have links to the old
+page, and can manually deal with them.
+
+In some cases, a redirection page will be wanted, to keep long-lived urls
+working. Since the meta plugin supports creating such pages, and since they
+won't always be needed, I think it will be simplest to just leave it up to
+the user to create such a redirection page after renaming a page.
+
+## who can delete/rename what?
+
+The source page must be editable by the user to be deleted/renamed.
+When renaming, the dest page must not already exist, and must be creatable
+by the user, too.
+
+lWhen deleting/renaming attachments, the `allowed_attachments` PageSpec
+is checked too.
+
+## RCS
+
+Three new functions are added to the RCS interface:
+
+* `rcs_remove(file)`
+* `rcs_rename(old, new)`
+* `rcs_commit_staged(message, user, ip)`
+
+See [[rcs_updates_needed_for_rename_and_remove]].
+
+## conflicts
+
+Cases to consider:
+
+* Alice clicks "delete" button for a page; Bob makes a modification;
+ Alice confirms deletion. Ideally in this case, Alice should get an error
+ message that there's a conflict.
+ Update: In my current code, alice's deletion will fail if the file was
+ moved or deleted in the meantime; if the file was modified since alice
+ clicked on the delete button, the modifications will be deleted too. I
+ think this is acceptable.
+* Alice opens edit UI for a page; Bob makes a modification; Alice
+ clicks delete button and confirms deletion. Again here, Alice should get
+ a conflict error. Note that this means that the rcstoken should be
+ recorded when the edit UI is first opened, not when the delete button is
+ hit.
+ Update: Again here, there's no conflict, but the delete succeeds. Again,
+ basically acceptible.
+* Alice and Bob both try to delete a page at the same time. It's fine for
+ the second one to get a message that it no longer exists. Or just to
+ silently fail to delete the deleted page..
+ Update: It will display an error to the second one that the page doesn't
+ exist.
+* Alice deletes a page; Bob had edit window open for it, and saves
+ it afterwards. I think that Bob should win in this case; Alice can always
+ notice the page has been added back, and delete it again.
+ Update: Bob wins.
+* Alice clicks "rename" button for a page; Bob makes a modification;
+ Alice confirms rename. This case seems easy, it should just rename the
+ modified page.
+ Update: it does
+* Alice opens edit UI for a page; Bob makes a modification; Alice
+ clicks rename button and confirms rename. Seems same as previous case.
+ Update: check
+* Alice and Bob both try to rename a page at the same time (to probably
+ different names). Or one tries to delete, and the other to rename.
+ I think it's acceptible for the second one to get an error message that
+ the page no longer exists.
+ Update: check, that happens
+* Alice renames a page; Bob had edit window open for it, and saves
+ it afterwards, under old name. I think it's acceptible for Bob to succeed
+ in saving it under the old name in this case, though not ideal.
+ Update: Behavior is the same as if Alice renamed the page and Bob created
+ a new page with the old name. Seems acceptable, though could be mildly
+ confusing to Bob (or Alice).
+* Alice starts creating a new page. In the meantime, Bob renames a
+ different page to that name. Alice should get an error message when
+ committing; and it should have conflict markers. Ie, this should work the
+ same as if Bob had edited the new page at the same time as Alice did.
+ Update: That should happen. Haven't tested this case yet to make sure.
+* Bob starts renaming a page. In the meantime, Alice creates a new page
+ with the name he's renaming it to. Here Bob should get a error message
+ that he can't rename the page to an existing name. (A conflict resolution
+ edit would also be ok.)
+ Update: Bob gets an error message.
+* Alice renames (or deletes) a page. In the meantime, Bob is uploading an
+ attachment to it, and finishes after the rename finishes. Is it
+ acceptible for the attachment to be saved under the old name?
+ Update: Meh. It's certianly not ideal; if Bob tries to save the page he
+ uploaded the attachment to, he'll get a message about it having been
+ deleted/renamed, and he can try to figure out what to do... :-/
+* I don't know if this is a conflict, but it is an important case to consider;
+ you need to make sure that there are no security holes. You dont want
+ someone to be able to rename something to /etc/passwd.
+ I think it would be enough that you cannot rename to a location outside
+ of srcdir, you cannot rename to a location that you wouldn't be able
+ to edit because it is locked, and you cannot rename to an existing page.
+
+ > Well, there are a few more cases (like not renaming to a pruned
+ > filename, and not renaming _from_ a file that is not a known source
+ > file or is locked), but yes, that's essentially it.
+ >
+ > PS, the first thing I do to any
+ > web form is type /etc/passwd and ../../../../etc/passwd into it. ;-) --[[Joey]]