Forward Catalyst subprocess diagnostics to Python

doc/releases/changelog-dev.md

@@ -19,6 +19,11 @@
   distribution over the branches. The counted resources are then scaled proportionally and summed.
   [(#3059)](https://github.com/PennyLaneAI/catalyst/pull/3059)
 
+* Warnings and diagnostics emitted by successful Catalyst compiler subprocesses are now forwarded to
+  Python callers instead of being silently discarded. LLVM diagnostic colors are preserved in
+  interactive terminals.
+  [(#3080)](https://github.com/PennyLaneAI/catalyst/pull/3080)
+
 * A new runtime transport layer for remote/local executors is introduced.
   [(#3043)](https://github.com/PennyLaneAI/catalyst/pull/3043)
   [(#3045)](https://github.com/PennyLaneAI/catalyst/pull/3045)

frontend/catalyst/compiler.py

@@ -302,27 +302,37 @@ def _get_catalyst_cli_cmd(*args, stdin=None):
     return cmd
 
 
-def _catalyst(*args, stdin=None, text=True):
+def _catalyst(*args, stdin=None, text=True, stderr_return=False):
     """Raw interface to catalyst
 
     echo ${stdin} | catalyst *args -
     catalyst *args
     """
     cmd = _get_catalyst_cli_cmd(*args, stdin=stdin)
+
+    if sys.stderr.isatty():
+        cmd.insert(1, "--color")
+
     try:
         result = subprocess.run(cmd, input=stdin, check=True, capture_output=True, text=text)
+
+        # Capture the diagnostic output from the compiler
+        if stderr_return and result.stderr:
+            stderr = result.stderr.decode() if isinstance(result.stderr, bytes) else result.stderr
+            print(stderr, end="", file=sys.stderr)
+
         return result.stdout
     except subprocess.CalledProcessError as e:
         raise CompileError(f"catalyst failed with error code {e.returncode}: {e.stderr}") from e
 
 
-def _quantum_opt(*args, stdin=None, text=True):
+def _quantum_opt(*args, stdin=None, text=True, stderr_return=False):
     """Raw interface to quantum-opt
 
     echo ${stdin} | catalyst --tool=opt *args -
     catalyst --tool=opt *args
     """
-    return _catalyst(("--tool", "opt"), *args, stdin=stdin, text=text)
+    return _catalyst(("--tool", "opt"), *args, stdin=stdin, text=text, stderr_return=stderr_return)
 
 
 def canonicalize(*args, stdin=None, options: Optional[CompileOptions] = None):
@@ -404,6 +414,7 @@ def to_mlir_opt(
     options: Optional[CompileOptions] = None,
     using_python_compiler=False,
     workspace=None,
+    stderr_return=True,
 ):
     """echo ${input} | catalyst --tool=opt *args *opts -"""
     # Check if we need to use the Python interface for xDSL passes
@@ -417,12 +428,12 @@ def to_mlir_opt(
 
     # These are the options that may affect compilation
     if not options:
-        return _quantum_opt(*args, stdin=stdin)
+        return _quantum_opt(*args, stdin=stdin, stderr_return=stderr_return)
 
     opts = _options_to_cli_flags(options)
     if workspace is not None:
         opts += [("--workspace", str(workspace))]
-    return _quantum_opt(*opts, *args, stdin=stdin)
+    return _quantum_opt(*opts, *args, stdin=stdin, stderr_return=stderr_return)
 
 
 class Compiler: