feat: add large stack test macro (#12768)

This PR adds the macro `#[large_stack_test]`

This spawns the tests in a dedicated tokio runtime with a larger stack.
It is useful for tests that needs the full recursion on the harness
(which is now too deep for windows for example)

codex-rs/Cargo.lock

@@ -1718,6 +1718,7 @@ dependencies = [
  "codex-shell-escalation",
  "codex-skills",
  "codex-state",
+ "codex-test-macros",
  "codex-utils-absolute-path",
  "codex-utils-cargo-bin",
  "codex-utils-home-dir",
@@ -2298,6 +2299,15 @@ dependencies = [
  "uds_windows",
 ]
 
+[[package]]
+name = "codex-test-macros"
+version = "0.0.0"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "codex-tui"
 version = "0.0.0"

codex-rs/Cargo.toml

@@ -62,6 +62,7 @@ members = [
     "codex-api",
     "state",
     "codex-experimental-api-macros",
+    "test-macros",
 ]
 resolver = "2"
 
@@ -116,6 +117,7 @@ codex-shell-command = { path = "shell-command" }
 codex-shell-escalation = { path = "shell-escalation" }
 codex-skills = { path = "skills" }
 codex-state = { path = "state" }
+codex-test-macros = { path = "test-macros" }
 codex-stdio-to-uds = { path = "stdio-to-uds" }
 codex-tui = { path = "tui" }
 codex-utils-absolute-path = { path = "utils/absolute-path" }

codex-rs/core/Cargo.toml

@@ -148,6 +148,7 @@ codex-arg0 = { workspace = true }
 codex-otel = { workspace = true, features = [
     "disable-default-metrics-exporter",
 ] }
+codex-test-macros = { workspace = true }
 codex-utils-cargo-bin = { workspace = true }
 core_test_support = { workspace = true }
 ctor = { workspace = true }

codex-rs/core/tests/suite/apply_patch_cli.rs

@@ -1,6 +1,7 @@
 #![allow(clippy::expect_used)]
 
 use anyhow::Result;
+use codex_test_macros::large_stack_test;
 use core_test_support::responses::ev_apply_patch_call;
 use core_test_support::responses::ev_apply_patch_custom_tool_call;
 use core_test_support::responses::ev_shell_command_call;
@@ -85,7 +86,7 @@ fn apply_patch_responses(
     ]
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -132,7 +133,7 @@ D delete.txt
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -159,7 +160,7 @@ async fn apply_patch_cli_multiple_chunks(model_output: ApplyPatchModelOutput) ->
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -188,7 +189,7 @@ async fn apply_patch_cli_moves_file_to_new_directory(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -216,7 +217,7 @@ async fn apply_patch_cli_updates_file_appends_trailing_newline(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -242,7 +243,7 @@ async fn apply_patch_cli_insert_only_hunk_modifies_file(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -273,7 +274,7 @@ async fn apply_patch_cli_move_overwrites_existing_destination(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -334,7 +335,7 @@ async fn apply_patch_cli_move_without_content_change_has_no_turn_diff(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -360,7 +361,7 @@ async fn apply_patch_cli_add_overwrites_existing_file(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -392,7 +393,7 @@ async fn apply_patch_cli_rejects_invalid_hunk_header(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -426,7 +427,7 @@ async fn apply_patch_cli_reports_missing_context(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -462,7 +463,7 @@ async fn apply_patch_cli_reports_missing_target_file(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -499,7 +500,7 @@ async fn apply_patch_cli_delete_missing_file_reports_error(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -524,7 +525,7 @@ async fn apply_patch_cli_rejects_empty_patch(model_output: ApplyPatchModelOutput
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -551,7 +552,7 @@ async fn apply_patch_cli_delete_directory_reports_verification_error(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -605,7 +606,7 @@ async fn apply_patch_cli_rejects_path_traversal_outside_workspace(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -660,7 +661,7 @@ async fn apply_patch_cli_rejects_move_path_traversal_outside_workspace(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -694,7 +695,7 @@ async fn apply_patch_cli_verification_failure_has_no_side_effects(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_shell_command_heredoc_with_cd_updates_relative_workdir() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -732,7 +733,7 @@ async fn apply_patch_shell_command_heredoc_with_cd_updates_relative_workdir() ->
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_cli_can_use_shell_command_output_as_patch_input() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -862,7 +863,7 @@ async fn apply_patch_cli_can_use_shell_command_output_as_patch_input() -> Result
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_shell_command_heredoc_with_cd_emits_turn_diff() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -945,7 +946,7 @@ async fn apply_patch_shell_command_heredoc_with_cd_emits_turn_diff() -> Result<(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_shell_command_failure_propagates_error_and_skips_diff() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -1021,7 +1022,7 @@ async fn apply_patch_shell_command_failure_propagates_error_and_skips_diff() ->
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::ShellViaHeredoc)]
 #[test_case(ApplyPatchModelOutput::ShellCommandViaHeredoc)]
 async fn apply_patch_function_accepts_lenient_heredoc_wrapped_patch(
@@ -1044,7 +1045,7 @@ async fn apply_patch_function_accepts_lenient_heredoc_wrapped_patch(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -1067,7 +1068,7 @@ async fn apply_patch_cli_end_of_file_anchor(model_output: ApplyPatchModelOutput)
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -1102,7 +1103,7 @@ async fn apply_patch_cli_missing_second_chunk_context_rejected(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -1161,7 +1162,7 @@ async fn apply_patch_emits_turn_diff_event_with_unified_diff(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]
@@ -1227,7 +1228,7 @@ async fn apply_patch_turn_diff_for_rename_with_content_change(
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_aggregates_diff_across_multiple_tool_calls() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -1295,7 +1296,7 @@ async fn apply_patch_aggregates_diff_across_multiple_tool_calls() -> Result<()>
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 async fn apply_patch_aggregates_diff_preserves_success_after_failure() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
@@ -1385,7 +1386,7 @@ async fn apply_patch_aggregates_diff_preserves_success_after_failure() -> Result
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[large_stack_test]
 #[test_case(ApplyPatchModelOutput::Freeform)]
 #[test_case(ApplyPatchModelOutput::Function)]
 #[test_case(ApplyPatchModelOutput::Shell)]

codex-rs/test-macros/BUILD.bazel

@@ -0,0 +1,7 @@
+load("//:defs.bzl", "codex_rust_crate")
+
+codex_rust_crate(
+    name = "test-macros",
+    crate_name = "codex_test_macros",
+    proc_macro = True,
+)

codex-rs/test-macros/Cargo.toml

@@ -0,0 +1,16 @@
+[package]
+name = "codex-test-macros"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[lib]
+proc-macro = true
+
+[dependencies]
+proc-macro2 = "1"
+quote = "1"
+syn = { version = "2", features = ["full"] }
+
+[lints]
+workspace = true

codex-rs/test-macros/src/lib.rs

@@ -0,0 +1,155 @@
+use proc_macro::TokenStream;
+use proc_macro2::TokenStream as TokenStream2;
+use quote::quote;
+use syn::Attribute;
+use syn::ItemFn;
+use syn::parse::Nothing;
+use syn::parse_macro_input;
+use syn::parse_quote;
+
+const LARGE_STACK_TEST_STACK_SIZE_BYTES: usize = 16 * 1024 * 1024;
+
+/// Run a test body on a dedicated thread with a larger stack.
+///
+/// For async tests, this macro creates a Tokio multi-thread runtime with two
+/// worker threads and blocks on the original async body inside the large-stack
+/// thread.
+#[proc_macro_attribute]
+pub fn large_stack_test(attr: TokenStream, item: TokenStream) -> TokenStream {
+    parse_macro_input!(attr as Nothing);
+
+    let item = parse_macro_input!(item as ItemFn);
+    expand_large_stack_test(item).into()
+}
+
+fn expand_large_stack_test(mut item: ItemFn) -> TokenStream2 {
+    let attrs = filtered_attributes(&item.attrs);
+    item.attrs = attrs;
+
+    let is_async = item.sig.asyncness.take().is_some();
+    let name = &item.sig.ident;
+    let body = &item.block;
+
+    let thread_body = if is_async {
+        quote! {
+            {
+                let runtime = ::tokio::runtime::Builder::new_multi_thread()
+                    .worker_threads(2)
+                    .enable_all()
+                    .build()
+                    .unwrap_or_else(|error| {
+                        panic!("failed to build tokio runtime for large-stack test: {error}")
+                    });
+                runtime.block_on(async move #body)
+            }
+        }
+    } else {
+        quote! { #body }
+    };
+
+    *item.block = parse_quote!({
+        let handle = ::std::thread::Builder::new()
+            .name(::std::string::String::from(::std::stringify!(#name)))
+            .stack_size(#LARGE_STACK_TEST_STACK_SIZE_BYTES)
+            .spawn(move || #thread_body)
+            .unwrap_or_else(|error| {
+                panic!("failed to spawn large-stack test thread: {error}")
+            });
+
+        match handle.join() {
+            Ok(result) => result,
+            Err(payload) => ::std::panic::resume_unwind(payload),
+        }
+    });
+
+    quote! { #item }
+}
+
+fn filtered_attributes(attrs: &[Attribute]) -> Vec<Attribute> {
+    let mut filtered = Vec::with_capacity(attrs.len() + 1);
+    let mut has_test_attr = false;
+
+    for attr in attrs {
+        if is_tokio_test_attr(attr) {
+            continue;
+        }
+        if is_test_attr(attr) || is_test_case_attr(attr) {
+            has_test_attr = true;
+        }
+        filtered.push(attr.clone());
+    }
+
+    if !has_test_attr {
+        filtered.push(parse_quote!(#[test]));
+    }
+
+    filtered
+}
+
+fn is_test_attr(attr: &Attribute) -> bool {
+    attr.path().is_ident("test")
+}
+
+fn is_test_case_attr(attr: &Attribute) -> bool {
+    attr.path().is_ident("test_case")
+}
+
+fn is_tokio_test_attr(attr: &Attribute) -> bool {
+    let mut segments = attr.path().segments.iter();
+    matches!(
+        (segments.next(), segments.next(), segments.next()),
+        (Some(first), Some(second), None) if first.ident == "tokio" && second.ident == "test"
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::expand_large_stack_test;
+    use syn::ItemFn;
+    use syn::parse_quote;
+
+    fn has_attr(item: &ItemFn, name: &str) -> bool {
+        item.attrs.iter().any(|attr| attr.path().is_ident(name))
+    }
+
+    #[test]
+    fn adds_test_attribute_when_missing() {
+        let item: ItemFn = parse_quote! {
+            fn sample() {}
+        };
+
+        let expanded_tokens = expand_large_stack_test(item);
+        let expanded: ItemFn = match syn::parse2(expanded_tokens) {
+            Ok(expanded) => expanded,
+            Err(error) => panic!("failed to parse expanded function: {error}"),
+        };
+
+        assert!(has_attr(&expanded, "test"));
+        let body = quote::quote!(#expanded).to_string();
+        assert!(body.contains("stack_size"));
+    }
+
+    #[test]
+    fn removes_tokio_test_and_keeps_test_case() {
+        let item: ItemFn = parse_quote! {
+            #[test_case(1)]
+            #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+            async fn sample(value: usize) -> anyhow::Result<()> {
+                let _ = value;
+                Ok(())
+            }
+        };
+
+        let expanded_tokens = expand_large_stack_test(item);
+        let expanded: ItemFn = match syn::parse2(expanded_tokens) {
+            Ok(expanded) => expanded,
+            Err(error) => panic!("failed to parse expanded function: {error}"),
+        };
+
+        assert!(has_attr(&expanded, "test_case"));
+        assert!(!has_attr(&expanded, "test"));
+        let body = quote::quote!(#expanded).to_string();
+        assert!(body.contains("tokio :: runtime :: Builder"));
+        assert!(!body.contains("tokio :: test"));
+    }
+}