the proposed patch creates a hook_delete_confirm, which is called from node_delete during the
confirmation cycle. this hook allows modules to inject custom control structures into the deletion
process, as they don't currently have this ability. this is needed, for example, in the
eventrepeat.module that i'm creating, which handles repeating/recurring events for event.module.
without a solution such as this, there is no way for the module to put in options for different
kinds of deletions--ex. 'this occurance only', 'this and all future occurances', 'all occurances'.
click here for a screenshot of what this hook looks like in action...
i've tested this patch and it works great, even when multiple modules are calling
hook_delete_confirm. i'd also be happy to write the documentation for hook_delete_confirm if this
patch gets accepted.
Comment | File | Size | Author |
---|---|---|---|
#39 | delete_pre_1.patch | 2.52 KB | hunmonk |
#34 | delete_pre_0.patch | 2.52 KB | hunmonk |
#33 | delete_pre.patch | 2.47 KB | hunmonk |
#32 | delete.pre.patch | 2.53 KB | Steven |
#23 | delete_confirm_doc_3.txt | 3.6 KB | hunmonk |
Comments
Comment #1
hunmonk CreditAttribution: hunmonk commentedjust a note that this patch will apply to both 4.6.x and HEAD. marking the version as 'cvs'. also, attached is the screenshot i mentioned above.
Comment #2
Steven CreditAttribution: Steven commenteda) New features don't get added to 4.6, this would only go into HEAD.
b) You're using the historic argument order to implode(), and leaving out the 2nd argument, which is not allowed according to the PHP docs. It is best to be precise: http://php.net/implode.
Comment #3
Steven CreditAttribution: Steven commentedAlso, you don't deal with mass-deletion. This needs to be added, otherwise things might break if your module depends on operations being performed on deletion.
Comment #4
hunmonk CreditAttribution: hunmonk commented>>a) New features don't get added to 4.6, this would only go into HEAD.
no problem there. since many people are waiting for this feature, i'll include this patch (or a patched version of node.module) so that the functionality will be available for those that wish to begin using my module right away in 4.6.x
>>b) You're using the historic argument order to implode(), and leaving out the 2nd argument, which is not >>allowed according to the PHP docs.
attached patch corrects this.
>>Also, you don't deal with mass-deletion. This needs to be added, otherwise things might break if your >>module depends on operations being performed on deletion.
this shouldn't be necessary, as the modules have ample tools to deal w/ mass deletion themselves--as long as they can pass some kind of control stucture into the deletion cycle (so that the module knows what kind of delete operation it needs to perform). i've tested the multiple delete functionality extensively in eventrepeat.module, and it works flawlessly with this patch in place. my thinking in designing this patch was add the absolute minimum to core that allows the modules some control in the deletion process. i think this is all that they'll need as long as they're well written.
Comment #5
Bèr Kessels CreditAttribution: Bèr Kessels commentedI do like the idea of this, since i ran into a similar issue for shazamgallery and for books.
But, steven is right when he says you should deal with batch deletion. in node/admin you can delete a range of nodes at once. your patch will break that.
And furthermore, I would like to see a good piece of documentation on how to use this. From what you have coded it seems the hook must return an array. Why not a straing? You concenate arrays with nothin ('') which IMHO is bad PHP. But I guess you have good reasons for returning a string.
Comment #6
hunmonk CreditAttribution: hunmonk commented>>But, steven is right when he says you should deal with batch deletion. in node/admin you can delete a >>range of nodes at once. your patch will break that.
not true. this patch isn't involved in the actual deletion of nodes, only with allowing modules to pass in a control structure so they have the ability to initiate different kinds of deletes--this doesn't interfere w/ the admin batch delete functionality at all. admin batch delete has it's own custom delete function, and this hook will not be called during it.
here is a detailed use case for this patch, so that the flow can be seen more clearly:
in eventrepeat.module, i have a need to be able to give the user a choice to delete entire repeat sequences, or just the current node. this needs to be decided at time of deletion, and i want to give them three choices:
'this occurance'
'this occurance and all future occurances' (which deletes from the date of the selected node forward)
'all occurances' (which deletes from today's date forward)
prior to this patch, there was no way that my module could inject this kind of control structure into the deletion process. here's the hook i put into eventrepeat.module to take advantage of the new feature provided by this patch:
function eventrepeat_delete_confirm($node) {
if (variable_get('event_nodeapi_'. $node->type, 'never') != 'never' &&
variable_get('eventrepeat_nodeapi_'. $node->type, 0) == 1) {
if ($node->eventrepeat_rid) {
$form .= form_hidden('eventrepeat_rid', $node->eventrepeat_rid);
$options = array(
'this' => t('This occurance only'),
'future' => t('This occurance and all future occurances'),
'all' => t('All occurances')
);
$form .= form_radios('Repeat event--delete the following:', 'eventrepeat_delete_type', 'this', $options, $description =
t('\'This occurance and all future occurances\' will delete repeat events from the date of the selected node forward,
\'All occurances\' will delete repeat events from today\'s date forward.'),
$required = FALSE, $attributes = NULL);
}
return $form;
}
}
this injects the id of the repeat sequence and the control structure i detailed above into the delete confirmation screen (i do some checks at the top to make sure that this data is only injected for nodes that are in a repeat sequence). now, when the user clicks 'Delete', the info i injected is available as $_POST data. since the user clicked delete, drupal processes the deletion of the node, and eventually during that process eventrepeat_nodeapi is called, at which point the module grabs the info it needs from $_POST. if, for example, the user selected 'all occurances', the module then calls it's own custom delete function to handle that process. simply put, the guts of that process is grabbing all the nids that need to be deleted, and calling node_delete for each one (in this case i pass in the confimation setting as well to node_delete, to automate the process). node_delete takes care of all the deletion dirty work for each node. the only tricky part is making sure that my custom delete function isn't called from eventrepeat_nodeapi multiple times (b/c that's called multiple times in a multiple delete)--i take care of this by unsetting the injected $_POST data when my custom delete function is called. works beautifully... :)
>>And furthermore, I would like to see a good piece of documentation on how to use this.
hopefully the above example makes usage clear, but i'd be happy to write up more doc if needed.
>>From what you have coded it seems the hook must return an array. Why not a straing?
the hook returns an html string which is the form data that will be injected into the confirmation screen
>>You concenate arrays with nothin ('') which IMHO is bad PHP. But I guess you have good reasons for >>returning a string.
these steps are necessary, because:
#1 module_invoke_all returns an array, and each element of the array is an html string that i described above. if multiple modules return form data to be injected into the confirmation screen, this array will have multiple elements
#2 the $extra argument in theme confirm needs to be passed in as a string, so i implode the array. it's best to include the 'glue' argument in implode (according to PHP documentation), but i don't want a seperator--i just want to turn the array into a string. therefore i insert an empty string as the 'glue'.
Comment #7
Bèr Kessels CreditAttribution: Bèr Kessels commented1: you are right about the batch deletion. In fact, that will even benefit from this, because it might be able to use this hook.
2: the fact that both Steven and I misread your patch shows that the use of this hook needs clarification. I am not sure how the hooks at drupaldocs are documented, but it really needs to show up there, with details about what to returns and what to pass along. Also some clarification about when this hook is called is important (for that caused the confusion at #1)
Comment #8
killes@www.drop.org CreditAttribution: killes@www.drop.org commentedI obviously support this patch. Aside from the demand for docs nothing stops it from being submitted to core. Since the docs for hook calls actually reside outside of core, even that is no reason to not commit it right now.
Comment #9
Bèr Kessels CreditAttribution: Bèr Kessels commentedoh, btw, you have my +1. forgot to add that in the previous issue. ;)
Comment #10
hunmonk CreditAttribution: hunmonk commented>>2: the fact that both Steven and I misread your patch shows that the use of this hook needs
>>clarification. I am not sure how the hooks at drupaldocs are documented, but it really needs to show up
>>there, with details about what to returns and what to pass along. Also some clarification about when
>>this hook is called is important (for that caused the confusion at #1)
i definitely agree. it's usage confused me in the beginning and i wrote it... ;)
if the patch gets committed, i'm happy to write supporting documentation for hook_delete_confirm and post it whereever it needs to go
Comment #11
hunmonk CreditAttribution: hunmonk commentedok, based on several discussions with core developers, i've expanded the functionality so that mods can inject info into the admin mass delete cycle as well. for single node deletions, the patch allows for multiple modules to inject info into the confirmation screen. for admin mass deletions the same capability exists, plus modules can also inject data for multiple nodes. for the admin part of the hook, i also cleaned up the code that prints the confirmation screen--removing some unnecessary database queries and putting the list of nodes to be deleted into a theme_item_list. because of the extra database queries i removed, i think performance will be about the same even with the new functionality. this patch introduces a very handy functionality, and i've already found use for it in two of the mods i'm developing... :) not only can it inject control structures to enable mods to handle different types of multiple deletions (which i use in eventrepeat.module), but it can also be used to inject custom module messages into the delete cycle (ex., in signup.module, i use it to warn users when they're about to delete a node that users are signed up for!)
for the module developer, this hook takes one argument, which is an array of nids to be deleted, and returns an array, the elements of which are also an array, with each key being the nid for which to inject the data, and the corresponding value is the data to be injected. (the array nesting was necessary because of the way module_invoke_all returns it's results. this approach returns each mod's custom injection data in it's own array--very flexible!)
you can see this hook in action here--username 'test' password 'drupal'. please also note that i didn't spend much time on working out the list theming, so we can clean that up later if it's not to everyone's liking!
Comment #12
hunmonk CreditAttribution: hunmonk commentedthat last link to the test site was no good.
try here--username 'test' password 'drupal'
also, i'll post some screenshots of the hook in action, and write up some better doc in the next day or two...
Comment #13
hunmonk CreditAttribution: hunmonk commentedattached are some screenshots of what this hook can do.
after some more talks w/ other developers, it seems as though this functionality may fit better in hook_nodeapi, as a 'delete confirm' case. this seems to make sense b/c all modules have the option to inject data into any of the nodes that are being deleted. i'm happy to modify the patch in this case, but there is one catch that i wanted to get some opinions on before i dive in...
in order for this hook to be supported in the admin batch delete, it's necessary to pass an array of nids to the module hook, so that the module can decide which nids to inject data into. the problem is, hook_nodeapi doesn't currently have an available argument where that can be passed in. what does everybody think about the idea of adding an $extra argument? the hook would then look like:
hook_nodeapi(&$node, $op, $teaser = NULL, $page = NULL, $extra = NULL)
Comment #14
hunmonk CreditAttribution: hunmonk commentedattached is module developer's documentation for this hook. If the code gets put into nodeapi as i mentioned above, i can modify this documentation accordingly.
Comment #15
hunmonk CreditAttribution: hunmonk commentedafter further irc discussions w/ chx, i've rewritten this hook to live in nodeapi, and added a really cool extra feature... :)
now modules can actually prevent deletion of a node by explicitly returning a value of FALSE. on a single node deletion, the user will be bounced back to the node/edit screen, and in batch deletions, the node will be removed from the batch delete queue prior to the confirmation screen. i left it up to module developers to drupal_set_message and tell their users what happened... :)
couple of things about this new approach:
1. it's less efficient than the old approach for batch deletions, because now node_load and node_invoke_nodeapi get called for each node in the batch, as opposed to the old way which just made one call for all nodes. however...
2. it now lives in nodeapi, where i believe it belongs, and
3. the implementation for module developers is now way, way easier and less complicated.
also, from an efficiency standpoint, i was able to completely remove all database queries from the confirm screen code in admin batch delete (the current approach queries the database once for each node in the batch, which i thought was inefficient). so i don't think the overall performance is going to be much different.
i'll post module developer code for this new version shortly. outside of the delete prevention functionality, the patch does the same thing as before--so the screenshots i posted above are still relevant.
Comment #16
hunmonk CreditAttribution: hunmonk commentednew module developer documentation for the nodeapi version of this patch attached.
Comment #17
chx CreditAttribution: chx commentedWhen a node calls node_view of other nodes, it's invaluable to be able to stop the deletion of its 'siblings'. I was doing evil hackery before: I str_replace'd in the theme to get rid of the delete button. So I like this feature. However, I am not fully convinced that full node object needs to be passed on during mass deletion. A simple
should be enough IMO. But then again, node_delete issues a node_load. For consistency reasons, maybe this should be changed. BTW. in HEAD, node_load is simply node_load($nid).
So what others think: is node_load necessary before deletion?
What about renaming the op to 'delete pre'?
Comment #18
chx CreditAttribution: chx commentedJust a note, because I love giving credits: the implementation of how deletions can be stopped is straight out from workflow.
Comment #19
Bèr Kessels CreditAttribution: Bèr Kessels commentedI beleive we need node_load. As Robin states: " i left it up to module developers to drupal_set_message and tell their users what happened" So it is a good thing to have teh complete node, including all the metadata around it, available.
Comment #20
hunmonk CreditAttribution: hunmonk commentedper chx's suggestions, changed node_load to node_load($nid), and renamed the nodeapi $op to 'delete pre'. also double-checked the coding style. unless anybody catches something else, i think this baby is ready to go!
Ber: btw, the name is Chad... :)
Comment #21
Bèr Kessels CreditAttribution: Bèr Kessels commentedSorry for the name, Chad. (what about two new profile fields: #drupal irc nick and REal name?)
Comment #22
hunmonk CreditAttribution: hunmonk commentedminor cleanup. just adding a comment line to clear up potential confusion--no functional changes
Comment #23
hunmonk CreditAttribution: hunmonk commented...and...
updated module developer doc for the latest version of the patch
the updated patch can be seen in action here
Comment #24
crunchywelch CreditAttribution: crunchywelch commented+1 from me, a lot of thought and discussion has happened around it, and I dont think there is another way to accomplish this without it. The patch seems small enough and will allow much needed functionality to progress. If only to quell the low roar of demand for repeating events, this patch should be committed...
Comment #25
matt westgate CreditAttribution: matt westgate commented+1
The ability to cancel the deletion of a node comes in handy for ecommerce when attempting to remove a still-valid recurring payment product that still has subscribers.
Comment #26
lopolencastredealmeida CreditAttribution: lopolencastredealmeida commented+1 from me too. Nothing to add to the discussion besides thanks.
Comment #27
hunmonk CreditAttribution: hunmonk commentedi've tested this patch extensively w/ no problems--works exactly as advertised. code has undergone review and several revisions, and is solid. developer feedback seems to indicate it's a desirable piece of functionality. switching to commit status...
Comment #28
bomarmonk CreditAttribution: bomarmonk commentedPlease get this in core for 4.7 (just adding my support, but I'm no developer--- just anxious for this functionality).
Comment #29
shane CreditAttribution: shane commentedI'd like to see this go into 4.7 so that repeating events can also go live. I've been needing repeating events for over 3 years - and have been manually adding repeating events every month.
HEARTY +1 for this.
Comment #30
benshell CreditAttribution: benshell commented+1 from me also. I'm also waiting for this before I upgrade one of my older (heavily modified) sites.
Comment #31
bkudrle CreditAttribution: bkudrle commented+1 for me also.
I am looking forward to the functionality provided by the recurring events.
Comment #32
Steven CreditAttribution: Steven commentedI was planning on committing this, but then I noticed that the mass delete changes are broken: the extra data in the mass delete form is never passed to node_delete(). I've attached a patch with cleaned up comments and code style, but the big is still there.
Comment #33
hunmonk CreditAttribution: hunmonk commentedCleaned up the pass by reference stuff, and doubled checked coding style. Mass delete changes should be fine--none of the module-specific data needs to be passed to node_delete, as the mods can grab whatever data that's posted from the confirm screen (which is the same way that node delete get it's confirmation info). Just to be clear, there are two steps in the deletion cycle, and the module developer can make use of this functionality in both steps:
Comment #34
hunmonk CreditAttribution: hunmonk commentedthat last patch seems to be malformed--here's a working one...
Comment #35
Dries CreditAttribution: Dries commentedI'm not a fan of this patch (as mentioned before). Deleting nodes is meant to be a simple task, not a programmable task. When I want to delete a node, I want to do exactly that: delete it.
Comment #36
decafdennis CreditAttribution: decafdennis commentedI don't know if I'm supposed to go against Dries and change the status back, but I do not completely agree with him.
I think he is referring to the fact that modules can prevent the deletion of a node. "When I want to delete a node, I want to do exactly that: delete it." Exactly. But...
That does not mean that the
delete pre
nodeapi operation can't be used to insert some extra information and/or form fields. I already have implemented this in one of my modules. This module provides some sort of container node, and I used thedelete pre
hook to add a checkbox to the delete confirmation form asking if the user wants to the delete all nodes that are attached to the container node. Works great.The cancellation thing can be removed from the patch. Please reconsider.
Comment #37
chx CreditAttribution: chx commentedReconsider indeed. I had my reasons for asking a feature so modules can turn down the deletion of a node. Matt provided a concrete use case http://drupal.org/node/26649#comment-40478 and also, node relation modules will need this.
Comment #38
decafdennis CreditAttribution: decafdennis commentedMaybe the cancellation code should be reworked, it is not very user-friendly atm, for example when deleting a single node the code simply redirects back to the edit page. A few suggestions are:
delete cancel
for example;Comment #39
hunmonk CreditAttribution: hunmonk commentedone more minor code change. it should be theme('item_list', $list), and not theme_item_list($list)...
Comment #40
chx CreditAttribution: chx commentedThe following criteria are used by core developers in reviewing and approving
proposed changes:
Comment #41
sepeck CreditAttribution: sepeck commentedI have to add my +1 voice here as well. I have one site that uses event module. Currently to have repeating or multi day classes, the store owner must hand enter each and every class time indivudally. This is a waste of her time if adding and removing multi day scheduled events can be done through one interface. The event repeat module worked on my test site with this patch in the brief time I had to test it.
She does not see 'nodes' as what she is deleting, she see's one class(event) that she is cancelling and deleting. That the event in question is actually occuring over 5 evening classes makes it 5 nodes confuses her, so this ability to automate node removal is important.
It is like when deleting a comment, all the child comments get deleted automatically too with but one warning.
Comment #42
Owen Barton CreditAttribution: Owen Barton commented+1 from me
Drupal is community plumbing.
To many communities events are the main form _of_ community (think a sports team).
Big communities have lots of events - these events repeat. It quickly becomes impossible to add all these events by hand. They also need to delete these events in a user friendly way.
Drupal needs to support this or it is not providing what these communities need!
Comment #43
hunmonk CreditAttribution: hunmonk commentedthis functionality will be included in the forthcoming undo functionality located here