arrow: add version 21.0.0

[amoeba]

Summary

Changes to recipe: arrow/21.0.0

Motivation

Updates recipe to use arrow 21.0.0, released July 17th, 2025. See https://github.com/apache/arrow/releases/tag/apache-arrow-21.0.0.

Details

Changes are scripted so this review process may be similar to previous ones.

---

• Read the contributing guidelines
• Checked that this PR is not a duplicate: list of PRs by recipe
• Tested locally with at least one configuration using a recent version of Conan

[franramirez688]

Thanks for this PR @amoeba 🙏

A couple of points before moving forward:

• As it was released yesterday, we tend to wait a few days in case a new patch is released.
• I think that we should apply the XXXXX-0001-fix-downloaded-mimalloc.patch (https://github.com/conan-io/conan-center-index/pull/27949/files#diff-e884b24b4d8cb748ee81015341d512175bbc07ebfa4cded266dd7b8c5d90f8e0L27-L30), but adapting the diff lines I guess.

[amoeba]

Thanks @franramirez688. Waiting a few days sounds fine and I'll integrate that patch later today.

[amoeba]

Sorry for the delay. I added the patch in 1c71513. The CI was complaining about RapidJSON which wasn't an expected failure.

[amoeba]

arrow 21.0.0 creates a new .so (built by default) so I'm expecting we'll also want a change like,

diff --git a/recipes/arrow/all/conanfile.py b/recipes/arrow/all/conanfile.py
index dfcc5707e..5d27b888a 100644
--- a/recipes/arrow/all/conanfile.py
+++ b/recipes/arrow/all/conanfile.py
@@ -446,6 +446,15 @@ class ArrowConan(ConanFile):
             self.cpp_info.components["libacero"].names["pkg_config"] = "acero"
             self.cpp_info.components["libacero"].requires = ["libarrow"]
 
+        if self.options.compute:
+            self.cpp_info.components["libcompute"].set_property("pkg_config_name", "compute")
+            self.cpp_info.components["libcompute"].set_property("cmake_target_name", f"Compute::arrow_compute_{cmake_suffix}")
+            self.cpp_info.components["libcompute"].libs = [f"arrow_compute{suffix}"]
+            self.cpp_info.components["libcompute"].names["cmake_find_package"] = "compute"
+            self.cpp_info.components["libcompute"].names["cmake_find_package_multi"] = "compute"
+            self.cpp_info.components["libcompute"].names["pkg_config"] = "compute"
+            self.cpp_info.components["libcompute"].requires = ["libarrow"]
+
         if self.options.gandiva:
             self.cpp_info.components["libgandiva"].set_property("pkg_config_name", "gandiva")
             self.cpp_info.components["libgandiva"].set_property("cmake_target_name", f"Gandiva::gandiva_{cmake_suffix}")

[franramirez688]

Hi @amoeba

I have just updated the patch with the correct lines (it was failing when applied) and the options, as parquet now requires RapidJSON.

Regarding the compute one, you're right. It's creating a new target now. I think that the changes to add should be something like:

        if self.options.compute:
            self.cpp_info.components["libarrow_compute"].set_property("pkg_config_name", "arrow_compute")
            self.cpp_info.components["libarrow_compute"].set_property("cmake_target_name", f"ArrowCompute::arrow_compute_{cmake_suffix}")
            self.cpp_info.components["libarrow_compute"].libs = [f"arrow_compute{suffix}"]
            self.cpp_info.components["libarrow_compute"].requires = ["libarrow"]

For the record (and not to be included in this PR), reviewing the rest of the components, I saw the existing one for acero, and I think it should be:

            self.cpp_info.components["libacero"].set_property("cmake_target_name", f"ArrowAcero::arrow_acero_{cmake_suffix}")

[amoeba]

Thanks for the catch on the acero thing, I think you're fix is right.

I pushed up a change to add libarrow_compute like you suggested.

Additionally, I think we need to specify that dataset now depends on compute, see https://github.com/apache/arrow/pull/46261/files#diff-1855e31fd38e7fa63c95344ad9992c27d2f609f015a06d45ca3dd3ecc708243d. I also think there may be a latent bug in that area of the conanfile because I think that libs should be additive and I think they're not currently. Here's my patch with both changes:

diff --git a/recipes/arrow/all/conanfile.py b/recipes/arrow/all/conanfile.py
index 222e71e21..af24dc7ef 100644
--- a/recipes/arrow/all/conanfile.py
+++ b/recipes/arrow/all/conanfile.py
@@ -477,11 +477,16 @@ class ArrowConan(ConanFile):
             self.cpp_info.components["libarrow_flight_sql"].requires = ["libarrow", "libarrow_flight"]
 
         if self.options.dataset_modules:
+            dataset_requires = []
             self.cpp_info.components["dataset"].libs = ["arrow_dataset"]
             if self.options.parquet:
-                self.cpp_info.components["dataset"].requires = ["libparquet"]
+                dataset_requires.append("libparquet")
             if self.options.acero and Version(self.version) >= "19.0.0":
-                self.cpp_info.components["dataset"].requires = ["libacero"]
+                dataset_requires.append("libacero")
+            if self.options.compute and Version(self.version) >= "21.0.0":
+                dataset_requires.append("libcompute")
+            self.cpp_info.components["dataset"].requires = dataset_requires
 
         if self.options.cli and (self.options.with_cuda or self.options.with_flight_rpc or self.options.parquet):
             binpath = os.path.join(self.package_folder, "bin")

cc @raulcd

[raulcd]

Acero depends on compute, and Dataset depends on Acero (and by transitive dependency on Compute), so I would add it as a requirement to Acero not to Compute.
Taking a look on the recipe there's also a couple of things that seems wrong, shouldn't the following be:

diff --git a/recipes/arrow/all/conanfile.py b/recipes/arrow/all/conanfile.py
index dfcc5707e..4dffde6a1 100644
--- a/recipes/arrow/all/conanfile.py
+++ b/recipes/arrow/all/conanfile.py
@@ -430,7 +430,7 @@ class ArrowConan(ConanFile):
 
         if self.options.get_safe("substrait"):
             self.cpp_info.components["libarrow_substrait"].set_property("pkg_config_name", "arrow_substrait")
-            self.cpp_info.components["libarrow_substrait"].set_property("cmake_target_name", f"Arrow::arrow_substrait_{cmake_suffix}")
+            self.cpp_info.components["libarrow_substrait"].set_property("cmake_target_name", f"ArrowSubstrait::arrow_substrait_{cmake_suffix}")
             self.cpp_info.components["libarrow_substrait"].libs = [f"arrow_substrait{suffix}"]
             self.cpp_info.components["libarrow_substrait"].requires = ["libparquet", "dataset"]
 
@@ -439,7 +439,7 @@ class ArrowConan(ConanFile):
 
         if self.options.acero:
             self.cpp_info.components["libacero"].set_property("pkg_config_name", "acero")
-            self.cpp_info.components["libacero"].set_property("cmake_target_name", f"Acero::arrow_acero_{cmake_suffix}")
+            self.cpp_info.components["libacero"].set_property("cmake_target_name", f"ArrowAcero::arrow_acero_{cmake_suffix}")
             self.cpp_info.components["libacero"].libs = [f"arrow_acero{suffix}"]
             self.cpp_info.components["libacero"].names["cmake_find_package"] = "acero"
             self.cpp_info.components["libacero"].names["cmake_find_package_multi"] = "acero"

Also, seems weird that dataset is treated differently as we can also expose the cmake_find_package, cmake_target_name as we do with others but we can do that on a separate PR. Link to documentation:
https://arrow.apache.org/docs/dev/cpp/build_system.html#other-available-packages

[franramirez688]

Thanks for the new feedback @amoeba @raulcd 👏

Then, let's address in this PR that requirement acero -> compute, could you @amoeba?
I'll open another PR later to fix all the mentioned stuff 😁

[amoeba]

Thanks @raulcd. Changed in e011d66.

[raulcd]

These changes LGTM, I understand we want to do other changes on a follow up PR.

[franramirez688]

LGTM! Thanks @amoeba @raulcd for the PR and the comments 👏

[amoeba]

Thanks @franramirez688 for the merge.