Implement `for f.(args...)` syntax by tkf · Pull Request #31553 · JuliaLang/julia

tkf · 2019-03-30T04:26:23Z

This PR implements for f.(args...) syntax as discussed in #19198 and planned in #31088 (comment).

close #19198
close #31088

Implementation detail

A new Expr(:fordot, :(broadcasted.(expression))) AST node is introduced to represent for broadcasted.(expression).
This is lowered to a chain of broadcasted calls by using the pre-existing code in julia-syntax.scm but skips the outer-most materialize call.
for non(dot(call)) is a syntax error at the moment.
Lowering manually constructed for non(dot(call)) is an error at the moment.
There is no performance specialization in this PR. It is done separately in Faster mapreduce for Broadcasted #31020.

Demo

julia> ex = :(for f.(x))
:(for f.(x))

julia> ex.head
:fordot

julia> ex.args
1-element Array{Any,1}:
 :(f.(x))

julia> Meta.lower(Main, ex)
:($(Expr(:thunk, CodeInfo(
    @ none within `top-level scope'
1 ─ %1 = Base.broadcasted(f, x)
└──      return %1
))))

julia> :(for f(x))
ERROR: syntax: invalid expression after `for`

julia> nondot = Expr(:fordot, :(f(x)))
:(for f(x))

julia> Meta.lower(Main, nondot)
:($(Expr(:error, "non-dot call after `for`")))

tkf · 2019-03-30T05:02:55Z

doc/src/devdocs/ast.md

 | `a==b`      | `(call == a b)`           |
 | `1<i<=n`    | `(comparison 1 < i <= n)` |
 | `a.b`       | `(. a (quote b))`         |
-| `a.(b)`     | `(. a b)`                 |


I removed a.(b) in "Operators" table (here) and mention it as f.(x) in "Calls" table since it seems the documentation does not match with the implementation:

> (julia-parse "a.(b)") (|.| a (tuple b))

stevengj · 2019-03-30T21:27:59Z

Does it support dot operators? e.g. for x .+ sqrt.(y)?

tkf · 2019-03-30T22:45:25Z

Yes

julia> x = .- (1:3)
-1:-1:-3

julia> y = x .^ 2
3-element Array{Int64,1}:
 1
 4
 9

julia> collect(for x .+ sqrt.(y))
3-element Array{Float64,1}:
 0.0
 0.0
 0.0

Its parsing and lowering are tested like

julia/test/syntax.jl

Lines 1846 to 1847 in 1882eae

    
           @test Meta.parse("for .+ x") == Expr(:fordot, :(.+ x)) 
        
           @test Meta.parse("for .+ g(x)") == Expr(:fordot, :(.+ g(x)))

and

julia/test/syntax.jl

Lines 1865 to 1874 in 1882eae

    
           @testset for valid in [ 
        
                   "for f.(x)" 
        
                   "for f.(g(x))" 
        
                   "for .+ x" 
        
                   "for .+ g(x)" 
        
                   "for f.(args[1], args[2])" 
        
                   "for f.(args...)" 
        
                   ] 
        
               @test Meta.lower(@__MODULE__, Meta.parse(valid)).head == :thunk 
        
           end

Let me know if there are other useful test cases.

chethega · 2019-03-31T17:27:45Z

So the correct use for iterating would be e.g. for i in for (0:10).^2 ... end resp for i = for (0:10).^2 ... end?

The doubling looks weirder than for i = @: (0:10).^2 ... end or for i = @lazy (0:10).^2 ... end.

antoine-levitt · 2019-03-31T18:09:17Z

I would imagine this pattern is less common than reducers (e.g. sum(for a.*b)): if you're going to iterate it, might as well do full iteration: for i=0:10 and then do something with i^2.

chethega · 2019-03-31T22:04:52Z

Fair enough. But Keno's original proposal in #31088 (comment) brought up exactly the point of lowering for i = (1:10).^2 body end into a non-materializing form that currently is spelled for i = for (0:10).^2 body end, at least if I understand this PR correctly.

tkf · 2019-03-31T23:32:57Z

a non-materializing form that currently is spelled for i = for (0:10).^2 body end, at least if I understand this PR correctly

@chethega Yes, that's correct. But this is not a new design. This is a consequence of what was proposed in the triage.

diegozea · 2019-04-04T00:32:42Z

I feel like it would be better/more explicit a @lazy macro or similar than adding new syntax to something so basic like for

tkf · 2019-04-04T00:47:15Z

I've implemented the macro-based solution in #31088 (I used different macro name but renaming it is quite easy). At this point I think it's up to triage to decide and merge the preferable one.

StefanKarpinski · 2019-04-04T05:44:42Z

I should mention that the for f.(args...) syntax was not universally loved and there was another syntax that got mentioned on another triage call that was a little strange at first but kind of makes sense: if we use &x to create a value reference and &v[i] to create an array reference and &x.f to create a field reference, then it could make sense to have &f(x) be a reference to the result of a function call, which would be lazy.

tkf · 2019-04-04T06:26:48Z

it could make sense to have &f(x) be a reference to the result of a function call

I think it does not work well if you want to denote Ref by & as well for the arguments of dot calls #27563 (comment). The tricky part here is that we have two "quasiquote/unquote-like" relationships: materialization/laziness and broadcasting/scalarization. Maybe it would be better to discuss #27563 and #19198 together?

adienes · 2025-12-24T19:21:52Z

given the staleness and the controversy about the design in the first place, I think it's fair to close this

tkf mentioned this pull request Mar 30, 2019

RFC: Use @: to construct a broadcasted object #31088

Closed

tkf added 2 commits March 29, 2019 21:36

Implement for f.(x) syntax

855373d

Add NEWS.md

a23e47b

tkf force-pushed the for-dot branch from 38a34a5 to a23e47b Compare March 30, 2019 04:36

tkf mentioned this pull request Mar 30, 2019

Notation for lazy map #19198

Open

Document new surface syntax in devdocs/ast.md

1882eae

tkf commented Mar 30, 2019

View reviewed changes

tkf mentioned this pull request Aug 3, 2019

What to do about nonscalar indexing? #30845

Open

tkf mentioned this pull request Aug 11, 2019

Julep: Syntax for reduction (overloadable splatting?) #32860

Open

tkf mentioned this pull request Feb 8, 2020

keys(::Generator) for find* and arg* #34674

Closed

baggepinnen mentioned this pull request Mar 5, 2020

Add function + two-argument method to reducers #35017

Closed

tkf mentioned this pull request Apr 10, 2020

searchsorted by applies by function to item being searched #9429

Open

adienes closed this Dec 24, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Implement `for f.(args...)` syntax#31553

Implement `for f.(args...)` syntax#31553
tkf wants to merge 3 commits intoJuliaLang:masterfrom
tkf:for-dot

tkf commented Mar 30, 2019 •

edited

Loading

Uh oh!

tkf Mar 30, 2019

Uh oh!

stevengj commented Mar 30, 2019

Uh oh!

tkf commented Mar 30, 2019 •

edited

Loading

Uh oh!

chethega commented Mar 31, 2019

Uh oh!

antoine-levitt commented Mar 31, 2019

Uh oh!

chethega commented Mar 31, 2019

Uh oh!

tkf commented Mar 31, 2019

Uh oh!

diegozea commented Apr 4, 2019

Uh oh!

tkf commented Apr 4, 2019

Uh oh!

StefanKarpinski commented Apr 4, 2019

Uh oh!

tkf commented Apr 4, 2019

Uh oh!

adienes commented Dec 24, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

Uh oh!

Conversation

tkf commented Mar 30, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Implementation detail

Demo

Uh oh!

tkf Mar 30, 2019

Choose a reason for hiding this comment

Uh oh!

stevengj commented Mar 30, 2019

Uh oh!

tkf commented Mar 30, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chethega commented Mar 31, 2019

Uh oh!

antoine-levitt commented Mar 31, 2019

Uh oh!

chethega commented Mar 31, 2019

Uh oh!

tkf commented Mar 31, 2019

Uh oh!

diegozea commented Apr 4, 2019

Uh oh!

tkf commented Apr 4, 2019

Uh oh!

StefanKarpinski commented Apr 4, 2019

Uh oh!

tkf commented Apr 4, 2019

Uh oh!

adienes commented Dec 24, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

tkf commented Mar 30, 2019 •

edited

Loading

tkf commented Mar 30, 2019 •

edited

Loading