https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123520
Bug ID: 123520
Summary: [Darwin] Master does not wait for dependent tasks
blocked on delay statement
Product: gcc
Version: 15.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: ada
Assignee: unassigned at gcc dot gnu.org
Reporter: mjgardner at abitofhelp dot com
CC: dkm at gcc dot gnu.org
Target Milestone: ---
Created attachment 63309
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=63309&action=edit
MacOS and Win11 example projects. .adb has commented results in each project.
On macOS/Darwin, a master subprogram does not wait for its dependent tasks to
terminate when those tasks are blocked on a delay statement. This violates ARM
9.3 which states the master cannot complete until all dependent tasks have
terminated.
The same code works correctly on Windows 11 with the same GNAT version.
Environment:
GNAT FSF 15.2.1 (via Alire)
GPRbuild 25.0.1
macOS Darwin 25.2.0 (Sequoia) - BUG
Windows 11 - WORKS CORRECTLY
Minimal reproducer:
with Ada.Text_IO; use Ada.Text_IO;
procedure Test is
task T;
task body T is
begin
for I in 1 .. 5 loop
Put_Line ("hello from task");
delay 1.0;
end loop;
Put_Line ("task done");
end T;
begin
Put_Line ("main");
end Test;
Expected behavior:
Program runs for ~5 seconds
Prints "hello from task" 5 times
Prints "task done"
Main waits at 'end Test' for task T to terminate
Actual behavior on macOS:
Program exits in ~0.3 seconds
Prints "hello from task" only 1 time
Never prints "task done"
Main does NOT wait for task T
Actual behavior on Windows:
Correct per Ada semantics
Workaround:
Use explicit entry/rendezvous synchronization
Note: If the delay statement is removed from the task body, macOS
behaves correctly. The bug only manifests when the task is blocked
on a delay statement at the time the master reaches its end.
