implement property destructuring #39285

simeonschaub · 2021-01-16T14:53:33Z

This currently only allows the form (; a, b) = x and lowers this to calls to getproperty. We could think about whether we want to allow specifying default values for properties as well, or even some kind of property renaming in the lhs. We could even allow slurping unused properties, but all of that sounds more difficult to work out in detail and potentially controversial, so I left this as an error for now.

fixes #28579

simeonschaub · 2021-01-16T14:54:58Z

src/julia-syntax.scm

+(define (expand-tuple-destruct lhss x)
+  (define (sides-match? l r)
+    ;; l and r either have equal lengths, or r has a trailing ...
+    (cond ((null? l)          (null? r))
+          ((vararg? (car l))  #t)
+          ((null? r)          #f)
+          ((vararg? (car r))  (null? (cdr r)))
+          (else               (sides-match? (cdr l) (cdr r)))))
+  (if (and (pair? x) (pair? lhss) (eq? (car x) 'tuple) (not (any assignment? (cdr x)))
+           (not (has-parameters? (cdr x)))
+           (sides-match? lhss (cdr x)))
+      ;; (a, b, ...) = (x, y, ...)
+      (expand-forms
+       (tuple-to-assignments lhss x))
+      ;; (a, b, ...) = other
+      (begin
+        ;; like memq, but if last element of lhss is (... sym),
+        ;; check against sym instead
+        (define (in-lhs? x lhss)
+          (if (null? lhss)
+              #f
+              (let ((l (car lhss)))
+                (cond ((and (pair? l) (eq? (car l) '|...|))
+                       (if (null? (cdr lhss))
+                           (eq? (cadr l) x)
+                           (error (string "invalid \"...\" on non-final assignment location \""
+                                          (cadr l) "\""))))
+                      ((eq? l x) #t)
+                      (else (in-lhs? x (cdr lhss)))))))
+        ;; in-lhs? also checks for invalid syntax, so always call it first
+        (let* ((xx  (if (or (and (not (in-lhs? x lhss)) (symbol? x))
+                            (ssavalue? x))
+                        x (make-ssavalue)))
+               (ini (if (eq? x xx) '() (list (sink-assignment xx (expand-forms x)))))
+               (n   (length lhss))
+               ;; skip last assignment if it is an all-underscore vararg
+               (n   (if (> n 0)
+                        (let ((l (last lhss)))
+                          (if (and (vararg? l) (underscore-symbol? (cadr l)))
+                              (- n 1)
+                              n))
+                        n))
+               (st  (gensy)))
+          `(block
+            ,@(if (> n 0) `((local ,st)) '())
+            ,@ini
+            ,@(map (lambda (i lhs)
+                     (expand-forms
+                       (if (vararg? lhs)
+                           `(= ,(cadr lhs) (call (top rest) ,xx ,@(if (eq? i 0) '() `(,st))))
+                           (lower-tuple-assignment
+                             (if (= i (- n 1))
+                                 (list lhs)
+                                 (list lhs st))
+                             `(call (top indexed_iterate)
+                                    ,xx ,(+ i 1) ,@(if (eq? i 0) '() `(,st)))))))
+                   (iota n)
+                   lhss)
+            (unnecessary ,xx))))))


NFC here, the level of indenting was just getting a bit much, so I moved this into a separate function.

CameronBieganek · 2021-01-16T15:44:26Z

Awesome! Thanks for this. Would you consider adding named tuple/struct destructuring in function arguments to this PR? That was the original ask in #28579.

simeonschaub · 2021-01-16T15:47:52Z

If you have a look at the tests, you can see that this already just works™️. Lowering works recursively, so we basically get this for free.

CameronBieganek · 2021-01-16T15:50:55Z

Nice! Yeah I was actually just looking at the tests now and noticed that. Thanks!

Do you know if the following syntax would work?

struct A
    x
    y
end

foo((; x, y)::A) = x + y

In other words, I want to be able to both dispatch on A and unpack the fields of A.

simeonschaub · 2021-01-16T15:52:50Z

Sure, that works just fine:

julia> struct A
           x
           y
       end

julia> foo((; x, y)::A) = x + y
foo (generic function with 1 method)

julia> foo(A(1, 2))
3

CameronBieganek · 2021-01-16T15:55:22Z

Ha, ok that example doesn't quite prove that it works fine. I need to see a method error if you call foo((x=1, y=2)). 😉

simeonschaub · 2021-01-16T16:02:28Z

Here you go 😄

julia> foo((x=1, y=2))
ERROR: MethodError: no method matching foo(::NamedTuple{(:x, :y), Tuple{Int64, Int64}})
Closest candidates are:
  foo(::A) at REPL[2]:1
Stacktrace:
 [1] top-level scope
   @ REPL[5]:1

If you don't want to build it yourself, you could also download the build from https://s3.amazonaws.com/julialangnightlies/assert_pretesting/linux/x64/1.7/julia-6aede6071e-linux64.tar.gz to try this out (assuming you are on Linux).

JeffBezanson · 2021-01-19T22:36:23Z

Nice! Seems elegant enough to me.

StefanKarpinski · 2021-01-19T22:44:22Z

Yes, this is a nice, elegant feature!

vtjnash · 2021-01-19T22:48:05Z

Agreed, I have wanted this! Though calling getindex rather than getproperty feels to me like it would mesh with more existing types and usages?

simeonschaub · 2021-01-19T23:42:43Z

Though calling getindex rather than getproperty feels to me like it would mesh with more existing types and usages?

One use case where I wanted something like this before was for destructuring DataFrames, which don't define getindex(x, ::Symbol). Besides a Dict with symbols as keys, I can't really think of anything that defines getindex(x, ::Symbol), but not getproperty. The nice thing about getproperty is also that it works with any user-defined type by default.

JeffBezanson · 2021-01-21T19:51:58Z

Triage is ok with this, but we note that it would be nice to add getproperty methods for Pairs of a NamedTuple, and really for that matter all AbstractDicts.

This currently only allows the form `(; a, b) = x` and lowers this to calls to `getproperty`. We could think about whether we want to allow specifying default values for properties as well, or even some kind of property renaming in the lhs. We could even allow slurping unused properties, but all of that sounds more difficult to work out in detail and potentially controversial, so I left this as an error for now. fixes #28579

simeonschaub · 2021-01-23T09:54:35Z

Buildbot failure is the usual "hard kill repl test".

* implement property destructuring This currently only allows the form `(; a, b) = x` and lowers this to calls to `getproperty`. We could think about whether we want to allow specifying default values for properties as well, or even some kind of property renaming in the lhs. We could even allow slurping unused properties, but all of that sounds more difficult to work out in detail and potentially controversial, so I left this as an error for now. fixes JuliaLang#28579 * add NEWS entry

BambOoxX · 2021-06-23T12:17:06Z

Sorry to barge in here with this, but after looking at the v1.7 announcement, I have a question.
I currently use Parameters.jl for this kind of things, with the @unpack macro, and I was wondering if the plan is to sort of integrate the functionalities of this package in Base, or if it is just sort of a coincidence or else ? Thanks.

simeonschaub · 2021-06-23T13:14:19Z

No, not really. Of course, if there is self-contained functionality that would make sense to have in Base, that can be discussed separately.

BambOoxX · 2021-06-25T12:26:33Z

Alright thanks !

stevengj · 2022-02-09T12:43:28Z

I can't find this documented in the manual anywhere?

Seems like an oversight in this PR?

fredrikekre · 2022-02-09T12:47:20Z

Docs where added in #41189

o314 · 2022-07-24T04:31:30Z

Hi. splat destructuring is not currently supported

julia> f((; head, arg)) = (; head, arg)
f (generic function with 1 method)

julia> f((; head, args...)) = (; head, args...)
ERROR: syntax: malformed expression

simeonschaub added compiler:lowering Syntax lowering (compiler front end, 2nd stage) needs news A NEWS entry is required for this change labels Jan 16, 2021

simeonschaub commented Jan 16, 2021

View reviewed changes

simeonschaub requested a review from JeffBezanson January 16, 2021 14:55

simeonschaub added the triage This should be discussed on a triage call label Jan 16, 2021

simeonschaub mentioned this pull request Jan 17, 2021

NamedTuple unpacking (and also struct unpacking) #28579

Closed

JeffBezanson approved these changes Jan 19, 2021

View reviewed changes

simeonschaub added 2 commits January 21, 2021 21:51

add NEWS entry

d2e2a9b

simeonschaub force-pushed the sds/property_destructuring branch from 6aede60 to d2e2a9b Compare January 21, 2021 20:51

fingolfin mentioned this pull request Jan 22, 2021

Fix jl_new_task to avoid allocating stack twice #39065

Merged

simeonschaub merged commit d9e2632 into master Jan 23, 2021

simeonschaub deleted the sds/property_destructuring branch January 23, 2021 09:55

simeonschaub removed needs news A NEWS entry is required for this change triage This should be discussed on a triage call labels Jan 23, 2021

aviatesk mentioned this pull request Jan 23, 2021

property destructuring doesn't allow lhs-type-declaration #39374

Closed

marius311 mentioned this pull request Jan 23, 2021

Match new Base destructuring syntax (aka the fate of this package) mauro3/UnPack.jl#14

Open

knuesel mentioned this pull request May 17, 2021

Combining pairs(::GroupDataFrame) JuliaData/DataFrames.jl#2593

Open

This was referenced Jul 13, 2021

New property destructuring (; a, b) = (a = 5, b = 6) doesn't work fonsp/Pluto.jl#1312

Closed

Allow assignment by destructuring fonsp/Pluto.jl#1317

Merged

henry2004y mentioned this pull request Sep 8, 2021

Built-in Julia 1.7 support for unpack henry2004y/Vlasiator.jl#26

Closed

simonbyrne mentioned this pull request Jan 10, 2022

WIP: build basic toml format CliMA/ClimaParams.jl#57

Closed

alfaromartino mentioned this pull request Oct 11, 2022

Deconstructing Named Tuples #47131

Closed

stevengj mentioned this pull request Oct 30, 2023

default values for destructuring syntax: (; x=default) = y #51940

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

implement property destructuring #39285

implement property destructuring #39285

simeonschaub commented Jan 16, 2021

simeonschaub Jan 16, 2021

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

JeffBezanson commented Jan 19, 2021

StefanKarpinski commented Jan 19, 2021

vtjnash commented Jan 19, 2021

simeonschaub commented Jan 19, 2021

JeffBezanson commented Jan 21, 2021

simeonschaub commented Jan 23, 2021

BambOoxX commented Jun 23, 2021

simeonschaub commented Jun 23, 2021

BambOoxX commented Jun 25, 2021

stevengj commented Feb 9, 2022

fredrikekre commented Feb 9, 2022

o314 commented Jul 24, 2022

implement property destructuring #39285

implement property destructuring #39285

Conversation

simeonschaub commented Jan 16, 2021

simeonschaub Jan 16, 2021

Choose a reason for hiding this comment

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

CameronBieganek commented Jan 16, 2021

simeonschaub commented Jan 16, 2021

JeffBezanson commented Jan 19, 2021

StefanKarpinski commented Jan 19, 2021

vtjnash commented Jan 19, 2021

simeonschaub commented Jan 19, 2021

JeffBezanson commented Jan 21, 2021

simeonschaub commented Jan 23, 2021

BambOoxX commented Jun 23, 2021

simeonschaub commented Jun 23, 2021

BambOoxX commented Jun 25, 2021

stevengj commented Feb 9, 2022

fredrikekre commented Feb 9, 2022

o314 commented Jul 24, 2022