Add optional argument to @inferred that relaxes the case of Nothing and Missing #27516

haampie · 2018-06-09T21:09:48Z

Brought up in #27511, where I want to test inference of an iterate call, yet @inferred is not applicable because the return type is Union{Nothing, T} by design.

The fix is to have @weakly_inferred, which does the same as @inferred, but relaxes the cases of nothing and missing:

> using Test
> v = [1, missing]
> @inferred v[1]
ERROR: return type Int64 does not match inferred return type Union{Missing, Int64}
> @weakly_inferred v[1]
1
> @weakly_inferred v[2]
missing
> @weakly_inferred rand([1, "str", nothing]) # note: might pass when `nothing` is returned!
ERROR: return type Int64 does not weakly match inferred return type Any

> using Test
> @inferred iterate(1:10)
ERROR: return type Tuple{Int64,Int64} does not match inferred return type Union{Nothing, Tuple{Int64,Int64}}
> @weakly_inferred iterate(1:10)
(1, 1)

If there's a better name for the macro, I'll be happy to change it.

Edit: final contribution in this PR is to add an optional type as argument to @inferred which is "white listed":

@inferred [Type] call()

nalimilan · 2018-06-10T09:47:19Z

Thanks! I agree it's useful, but would a syntax like @inferred Union{T,Missing} f(x) work instead?

haampie · 2018-06-10T10:49:48Z

Yeah, I think extending @inferred is better. What about @inferred f(x) T where the test passes if return_type <: T || return_type == typesubtract(inferred_type, T)?

@inferred iterate(...) Nothing
@inferred getdata(...) Missing
@inferred anything(...) Union{Nothing,Missing}

StefanKarpinski · 2018-06-10T16:15:37Z

Putting the type argument first seems more legible and intuitive to me.

martinholters · 2018-06-10T17:54:20Z

If you need to explicitly give the expected return type, what's the advantage compared to a type assertion?

haampie · 2018-06-10T18:11:50Z

If you need to explicitly give the expected return type, what's the advantage compared to a type assertion?

You don't know what the inferred type minus the expected / allowed return type is. For example:

function iterate(it, state)
   some_condition && return nothing
   new_value = ...
   new_state = ...
   new_value, new_state
end

I know iterate(it, state) gets inferred to Union{Nothing, T} for some T I am not sure about, so I want to test whether a concrete value y = iterate(it, state) has type T (when it is not nothing).

With the old behaviour of @inferred you could test it by splitting the iterate function like this:

function produce_new_state_and_value(it, state)
  new_value = ...
  new_state = ...
  new_value, new_state
end

function iterate(it, state)
  some_condition && return nothing
  produce_new_state_and_value(it, state)
end

@inferred produce_new_state_and_value(someiterator, somestate)

but often these functions are just 3 lines long, and it wouldn't make sense to define a new function just to test inference. With this PR, you would just run

@inferred Nothing iterate(someiterator, somestate)

I think this test is very useful now that Union{Nothing, T} and Union{Missing, T} are not really considered type instabilities anymore.

ararslan · 2018-06-11T07:22:49Z

stdlib/Test/src/Test.jl

@@ -21,7 +21,7 @@ export @test, @test_throws, @test_broken, @test_skip,

 export @testset
 # Legacy approximate testing functions, yet to be included
-export @inferred
+export @inferred, @weakly_inferred


This should be removed since you're no longer defining it.

martinholters · 2018-06-11T08:14:52Z

Right, sorry, I had misinterpreted the proposal here. Carry on.

haampie · 2018-06-14T06:34:15Z

Is this ready to be merged? Then I could use it in #27511 :)

ararslan · 2018-06-14T18:28:51Z

Yeah, looks good to me. Though perhaps we should rerun CI since it's been a couple days?

haampie · 2018-06-15T16:56:55Z

Circleci is killed during the build, and Travis probably the same? Other look good.

ararslan · 2018-06-15T19:16:53Z

The CI failures are doctest failures.

haampie · 2018-06-22T17:38:51Z

Fixed the doctest issue, things look good. Not sure why circleci is failing.

nalimilan · 2018-12-21T20:47:04Z

Too bad this wasn't merged. Could you rebase it?

haampie · 2018-12-21T21:10:52Z

Sure

haampie · 2018-12-22T00:26:15Z

Error has to do with the maximum PR?

Error in testset Profile:
Test Failed at C:\projects\julia\julia-\share\julia\stdlib\v1.2\Profile\test\runtests.jl:34
  Expression: !(isempty(str))
Error in testset Profile:
Test Failed at C:\projects\julia\julia-\share\julia\stdlib\v1.2\Profile\test\runtests.jl:41
  Expression: !(isempty(String(take!(iobuf))))
Error in testset Profile:
Error During Test at C:\projects\julia\julia-\share\julia\test\testdefs.jl:20
  Got exception outside of a @test
  LoadError: ArgumentError: reducing over an empty collection is not allowed
  Stacktrace:
   [1] _empty_reduce_error() at .\reduce.jl:216
   [2] reduce_empty(::Function, ::Type) at .\reduce.jl:226
   [3] mapreduce_empty(::typeof(identity), ::Function, ::Type) at .\reduce.jl:251
   [4] _mapreduce at .\reduce.jl:305 [inlined]
   [5] _mapreduce_dim at .\reducedim.jl:308 [inlined]
   [6] #mapreduce#549 at .\reducedim.jl:304 [inlined]
   [7] mapreduce at .\reducedim.jl:304 [inlined]
   [8] _maximum at .\reducedim.jl:653 [inlined]
   [9] _maximum at .\reducedim.jl:652 [inlined]
   [10] #maximum#555 at .\reducedim.jl:648 [inlined]
   [11] maximum at .\reducedim.jl:648 [inlined]
   [12] print_flat(::Base.GenericIOBuffer{Array{UInt8,1}}, ::Array{Base.StackTraces.StackFrame,1}, ::Array{Int32,1}, ::Int32, ::Profile.ProfileFormat) at C:\projects\julia\usr\share\julia\stdlib\v1.2\Profile\src\Profile.jl:398
   [13] flat(::Base.GenericIOBuffer{Array{UInt8,1}}, ::Array{UInt64,1}, ::Dict{UInt64,Array{Base.StackTraces.StackFrame,1}}, ::Int32, ::Profile.ProfileFormat) at C:\projects\julia\usr\share\julia\stdlib\v1.2\Profile\src\Profile.jl:368
   [14] print(::Base.GenericIOBuffer{Array{UInt8,1}}, ::Array{UInt32,1}, ::Dict{UInt64,Array{Base.StackTraces.StackFrame,1}}, ::Profile.ProfileFormat, ::Symbol) at C:\projects\julia\usr\share\julia\stdlib\v1.2\Profile\src\Profile.jl:149
   [15] (::getfield(Profile, Symbol("#kw##print")))(::NamedTuple{(:format, :sortedby),Tuple{Symbol,Symbol}}, ::typeof(Profile.print), ::Base.GenericIOBuffer{Array{UInt8,1}}) at C:\projects\julia\usr\share\julia\stdlib\v1.2\Profile\src\Profile.jl:134
   [16] top-level scope at C:\projects\julia\julia-\share\julia\stdlib\v1.2\Profile\test\runtests.jl:43

nalimilan · 2018-12-22T11:31:39Z

Looks like problem with stack traces (probably not related)... :-/

haampie · 2018-12-23T15:19:04Z

Tests are green now @nalimilan

KristofferC · 2019-01-02T11:50:31Z

Would be good with a news entry so people can more easily discover this :)

Julia 1.0's cumsum doesn't work for iterators -> Collect first @inferred didn't support Allowed types until v1.2 - Added in JuliaLang/julia#27516 - Add VERSION check and only run type stability checks if >= v1.2 .≈ does seem to be defined in v1.0 (Weird) - Use separate Eq Checks if 1D or 2D (Don't need to map at all if 1D) - Replace broadcasting with map -> Seems to work

haampie changed the title ~~[RFC] Add @inferred-like test macro that passes for functions returning nothing or missing~~ [RFC] Add @inferred-like test macro that relaxes the case of Nothing and Missing Jun 9, 2018

haampie force-pushed the feature-test-inference-with-small-unions branch from a1fdab0 to a3294e6 Compare June 10, 2018 07:55

haampie force-pushed the feature-test-inference-with-small-unions branch from a3294e6 to 1263dd4 Compare June 10, 2018 13:56

ararslan reviewed Jun 11, 2018

View reviewed changes

ararslan approved these changes Jun 11, 2018

View reviewed changes

haampie force-pushed the feature-test-inference-with-small-unions branch from 047b715 to dec2417 Compare June 14, 2018 20:08

haampie force-pushed the feature-test-inference-with-small-unions branch from dec2417 to 3379a94 Compare June 21, 2018 21:55

haampie mentioned this pull request Jun 24, 2018

Simplify zip take 2 #27511

Closed

haampie mentioned this pull request Oct 4, 2018

Change Zip data structure to a flat one #29238

Merged

haampie force-pushed the feature-test-inference-with-small-unions branch 3 times, most recently from e36a2e4 to 2264cdb Compare October 5, 2018 10:37

haampie changed the title ~~[RFC] Add @inferred-like test macro that relaxes the case of Nothing and Missing~~ Add optional argument to @inferred that relaxes the case of Nothing and Missing Oct 5, 2018

haampie force-pushed the feature-test-inference-with-small-unions branch from 2264cdb to 01d8124 Compare December 21, 2018 21:10

haampie force-pushed the feature-test-inference-with-small-unions branch from 01d8124 to fa19651 Compare December 21, 2018 21:12

Add test macro for relaxed inference of small unions

a50a8ca

haampie force-pushed the feature-test-inference-with-small-unions branch from fa19651 to a50a8ca Compare December 21, 2018 21:17

haampie closed this Dec 23, 2018

haampie reopened this Dec 23, 2018

andreasnoack merged commit f713a4b into JuliaLang:master Jan 2, 2019

KristofferC added the needs news A NEWS entry is required for this change label Jan 2, 2019

fchorney mentioned this pull request Jun 10, 2020

Add 2 arg @inferred JuliaLang/Compat.jl#706

Merged

JeffBezanson mentioned this pull request Feb 2, 2022

behavior of @inferred #44025

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add optional argument to @inferred that relaxes the case of Nothing and Missing #27516

Add optional argument to @inferred that relaxes the case of Nothing and Missing #27516

haampie commented Jun 9, 2018 •

edited

Loading

nalimilan commented Jun 10, 2018

haampie commented Jun 10, 2018

StefanKarpinski commented Jun 10, 2018

martinholters commented Jun 10, 2018

haampie commented Jun 10, 2018 •

edited

Loading

ararslan Jun 11, 2018

haampie Jun 11, 2018

martinholters commented Jun 11, 2018

haampie commented Jun 14, 2018

ararslan commented Jun 14, 2018

haampie commented Jun 15, 2018

ararslan commented Jun 15, 2018

haampie commented Jun 22, 2018

nalimilan commented Dec 21, 2018

haampie commented Dec 21, 2018

haampie commented Dec 22, 2018

nalimilan commented Dec 22, 2018

haampie commented Dec 23, 2018

KristofferC commented Jan 2, 2019

Add optional argument to @inferred that relaxes the case of Nothing and Missing #27516

Add optional argument to @inferred that relaxes the case of Nothing and Missing #27516

Conversation

haampie commented Jun 9, 2018 • edited Loading

nalimilan commented Jun 10, 2018

haampie commented Jun 10, 2018

StefanKarpinski commented Jun 10, 2018

martinholters commented Jun 10, 2018

haampie commented Jun 10, 2018 • edited Loading

ararslan Jun 11, 2018

Choose a reason for hiding this comment

haampie Jun 11, 2018

Choose a reason for hiding this comment

martinholters commented Jun 11, 2018

haampie commented Jun 14, 2018

ararslan commented Jun 14, 2018

haampie commented Jun 15, 2018

ararslan commented Jun 15, 2018

haampie commented Jun 22, 2018

nalimilan commented Dec 21, 2018

haampie commented Dec 21, 2018

haampie commented Dec 22, 2018

nalimilan commented Dec 22, 2018

haampie commented Dec 23, 2018

KristofferC commented Jan 2, 2019

haampie commented Jun 9, 2018 •

edited

Loading

haampie commented Jun 10, 2018 •

edited

Loading