commit 64fc730f767de84835a5f1b4fc9b7831a3c2d15b
author Romain Poncet <romain.pct@gmail.com> Sun May 05 20:07:12 2024 +0100
committer GitHub <noreply@github.com> Sun May 05 15:07:12 2024 -0400
tree f8d411d1b45459368567dc51f683ed78a919d795
parent ac199b0e377502ea0f1fa5ced7fda897a01b82a9

Compatibility of schemas with nested types (#504)

Hi,

Here is a minimal example of the issue I've encountered.

```julia
 struct A
    x::Int
end

struct B
    a::A
end

v = [B(A(i)) for i =1:3]

io = IOBuffer()
Arrow.write(io, v; file=false)
seekstart(io)
Arrow.append(io, v) # throws
```

I don't know if this is really necessary, or if I'm not using this
library properly, but this issue makes it difficult to append to arrow
files with nested types.

Since I've only added more cases where the call to `append` can succeed,
I do not think that this creates retro-compatibility issues.

Thanks for the review!

---------

Co-authored-by: Ben Baumgold <4933671+baumgold@users.noreply.github.com>

Project.toml

diff --git a/Project.toml b/Project.toml
index b283dc9..8f0cdf7 100644
--- a/Project.toml
+++ b/Project.toml
@@ -17,7 +17,7 @@
 name = "Arrow"
 uuid = "69666777-d1a9-59fb-9406-91d4454c9d45"
 authors = ["quinnj <quinn.jacobd@gmail.com>"]
-version = "2.7.1"
+version = "2.7.2"
 
 [deps]
 ArrowTypes = "31f734f8-188a-4ce0-8406-c8a06bd891cd"

src/append.jl

diff --git a/src/append.jl b/src/append.jl
index 5a9c259..64a638b 100644
--- a/src/append.jl
+++ b/src/append.jl
@@ -282,9 +282,31 @@
     for (t1, t2) in zip(sch1.types, sch2.types)
         tt1 = Base.nonmissingtype(t1)
         tt2 = Base.nonmissingtype(t2)
-        if t1 == t2 ||
-           (tt1 <: AbstractVector && tt2 <: AbstractVector && eltype(tt1) == eltype(tt2))
+        if t1 == t2
             continue
+        elseif tt1 <: AbstractVector && tt2 <: AbstractVector && eltype(tt1) == eltype(tt2)
+            continue
+        elseif isstructtype(tt1) && isstructtype(tt2)
+            is_equivalent_type_by_field(tt1, tt2)
+        else
+            return false
+        end
+    end
+    true
+end
+
+function is_equivalent_type_by_field(T1, T2)
+    n1 = fieldcount(T1)
+    n2 = fieldcount(T2)
+    n1 != n2 && return false
+
+    for i = 1:n1
+        fieldname(T1, i) == fieldname(T2, i) || return false
+
+        if fieldtype(T1, i) == fieldtype(T2, i)
+            continue
+        elseif isstructtype(T1) && isstructtype(T2)
+            is_equivalent_type_by_field(T1, T2) || continue
         else
             return false
         end

test/runtests.jl

diff --git a/test/runtests.jl b/test/runtests.jl
index ed288b3..6cf990a 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -1042,5 +1042,24 @@
                 @test tbl.f[2] === Foo493(4, 5)
             end
         end
+
+        @testset "# 504" begin
+            struct Foo504
+                x::Int
+            end
+
+            struct Bar504
+                a::Foo504
+            end
+
+            v = [Bar504(Foo504(i)) for i = 1:3]
+            io = IOBuffer()
+            Arrow.write(io, v; file=false)
+            seekstart(io)
+            Arrow.append(io, v) # testing the compatility between the schema of the arrow Table, and the "schema" of v (using the fallback mechanism of Tables.jl)
+            seekstart(io)
+            t = Arrow.Table(io)
+            @test Arrow.Tables.rowcount(t) == 6
+        end
     end # @testset "misc"
 end