<< Back to previous view

[CLJ-1251] The update function: like update-in, for first level Created: 03/Sep/13  Updated: 23/May/14

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.6
Fix Version/s: Release 1.7

Type: Enhancement Priority: Minor
Reporter: Michael O. Church Assignee: Ambrose Bonnaire-Sergeant
Resolution: Unresolved Votes: 3
Labels: None

Attachments: Text File CLJ-1251.patch     Text File update.patch    
Patch: Code and Test
Approval: Screened


update-in is useful for updating nested structures. Very often we just want to update one level, so an update function optimised for this use case is useful.

It operates identically to update-in with a key path of length one so these are the same:

(update-in m [k] f args...)
(update m k f args...)

Patch: CLJ-1251.patch

Screened by: Alex Miller

Comment by Alex Miller [ 06/Sep/13 9:56 AM ]

I like this - kind of halfway between assoc and update-in.

Comment by Michael O. Church [ 07/Sep/13 12:41 PM ]

It's very useful. I assumed that its non-inclusion was for a reason (hence was hesitant to submit the patch) but it comes in handy a lot. One project I'd like to do with some free time is a library for turn-based strategy games, which use update frequently to express game-state changes.

The downside of this change is that 'update is probably a defined function in a good number of modules written by other people. IMO the strongest reason not to include it is that it's such a common name; but the benefits (in my view) outweigh the downsides.

Comment by Andy Fingerhut [ 14/Feb/14 11:50 AM ]

Patch update.patch dated Sep 3 2013 no longer applies cleanly to latest Clojure master as of Feb 14 2014. It did on Feb 7 2014. I haven't checked in detail, but this is probably simply due to some tests recently added to a test file that require updating some diff context lines.

Comment by Ambrose Bonnaire-Sergeant [ 06/May/14 2:36 PM ]

The vararg validation should be done in the same way as `assoc`.

Comment by Alan Malloy [ 06/May/14 2:41 PM ]

The most obvious reason, to me, that clojure.core/update doesn't exist already is that it's not clear what it should do when given more than 3 arguments. Consider, for example, (update m a b c d). What does this do? There are at least three reasonable interpretations: (update-in m [a] b c d), passing c and d as extra args to the function b; (-> m (update-in [a] b) (update-in [c] d)), treating the args as alternating key/function pairs; (reduce (fn [m k] (update-in m [k] a)) m [b c d]), treating a as a function to apply to each of b, c, and d.

Any of these are plausible meanings for the vague name "update", and there's no obvious behavior to choose, whereas there's only one reasonable way for assoc and assoc-in to behave. If one of them were chosen, it would be a little bit nontrivial to read code using it, at least until it became so well-known that everyone thinks it's obvious. I don't have anything against this function that Michael Church has written, or including it in core, but I don't like naming it update, as if it were the only possible dual to update-in.

Comment by Kyle Kingsbury [ 06/May/14 4:09 PM ]

I'd like to second Alan Malloy's concern; I've defined (update m k f arg1 arg2) in most of my Clojure work to be "change the value for this key to be (f current-value arg1 arg2 ...)"; this is consistent with swap!, update-in, etc., and is in my experience the most common need for update. It also composes well with swap! and other higher-order friends. I suggest we use that variant instead, and rely on assoc or -> threading when updating multiple fields.

Comment by Michael O. Church [ 07/May/14 10:32 AM ]

I agree with Kyle and Alan. There are several interpretations of how update should behave and while it's not clear which one is "correct", Kyle's is most consistent with the rest of the language and therefore probably more right than the one I started with.

The issue I see with including an "update" function is that it will break code for others who've defined it for themselves. Kyle's interpretation is more consistent with the rest of Clojure and will probably involve the least breakage. I'd be happy using his version, and renaming mine to something else.

Comment by Rich Hickey [ 13/May/14 6:09 AM ]

I am in favor, and it should work like everything else: (update m k f args...)

Comment by Ambrose Bonnaire-Sergeant [ 13/May/14 7:18 AM ]

I'm working on a new patch.

Comment by Ambrose Bonnaire-Sergeant [ 13/May/14 7:39 AM ]

update-like-update-in.patch is the new patch as Rich requests.

Comment by Alex Miller [ 13/May/14 8:56 AM ]

Ambrose, I think the example in the description no longer follows the (update m k f args...) form right? Can you update?

Comment by Ambrose Bonnaire-Sergeant [ 13/May/14 9:46 AM ]

Alex, I'm not sure what you're referencing?

Comment by Ambrose Bonnaire-Sergeant [ 13/May/14 9:47 AM ]

If you mean the docstring, I did try and update it for update by copying update-in and change and plural keys to singular.

Comment by Alex Miller [ 13/May/14 10:18 AM ]

I mean the description for this ticket needs to be updated to reflect what we are currently considering.

Comment by Alex Miller [ 13/May/14 12:57 PM ]

In the patch, the docstring has "If the key does not exist, a hash-map will be created." which is not applicable in update right? I think it would be more accurate to say that the fn will be invoked on nil.

This line occurs twice in the tests:

{:a [1 2]}   (update {:a [1]} :a conj 2)

There is no test for what happens when the key is absent. For example:

(update {:a 1} :b str)
=> {:b "", :a 1}
Comment by Ambrose Bonnaire-Sergeant [ 13/May/14 1:30 PM ]

I removed the mention of creating hash-maps, and replaced it with the explicit behaviour of passing `nil` for missing keys.

FWIW I proposed a similar wording in the patch for http://dev.clojure.org/jira/browse/CLJ-373

Added a test for missing key. Removed the duplicate test.

Comment by Gary Fredericks [ 16/May/14 8:45 PM ]

Is it worth unrolling several arities for the sake of premature optimization? e.g., https://github.com/Prismatic/plumbing/blob/master/src/plumbing/core.clj#L33-41

Comment by Alex Miller [ 22/May/14 8:14 AM ]

I think that's probably worth doing - who can update the patch with multiple arities?

Comment by Alex Miller [ 23/May/14 11:25 AM ]

Ambrose, can you (or anyone else really) update the patch to unroll small arities?

Comment by Ambrose Bonnaire-Sergeant [ 23/May/14 11:40 AM ]

Yes will do now.

Comment by Ambrose Bonnaire-Sergeant [ 23/May/14 12:16 PM ]

Add multiple arities + tests (CLJ-1251.patch)

Generated at Wed Jul 23 10:37:49 CDT 2014 using JIRA 4.4#649-r158309.