error : use of undeclared identifier 'oparg' when building the tier 2 interpreter with tail calling

[winapiadmin]

Bug report

Bug description:

C:\cpython\PCbuild\pythoncore.vcxproj(720,5): warning : Toolset ClangCL is not used for official builds. Your build may have errors or incompatibilities.
..\Modules\socketmodule.h(21,11): warning : the current #pragma pack alignment value is modified in the included file [-Wpragma-pack] [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Modules\socketmodule.h(53,11): warning : the current #pragma pack alignment value is modified in the included file [-Wpragma-pack] [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\generated_cases.c.h(5584,13): error : use of undeclared identifier 'next_uop' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\generated_cases.c.h(5584,13): error : use of undeclared label 'enter_tier_two' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\generated_cases.c.h(7836,25): error : use of undeclared identifier 'next_uop' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\generated_cases.c.h(7836,25): error : use of undeclared label 'enter_tier_two' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(30,13): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(31,18): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(79,13): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(80,44): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(85,82): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(99,13): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(102,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(102,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(102,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(102,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(111,13): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(114,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(114,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(114,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
..\Python\executor_cases.c.h(114,45): error : use of undeclared identifier 'oparg' [C:\cpython\PCbuild\pythoncore.vcxproj]
TRACKEDEXEC : fatal error : too many errors emitted, stopping now [-ferror-limit=] [C:\cpython\PCbuild\pythoncore.vcxproj]
    3 Warning(s)
    20 Error(s)

(with tail call optimization, edit: + JIT tier 2)

But tail call optimization is supported on LLVM. I think this shouldn't be a duplicate of #141937

Related: #132132

(using both MSVC's clang-cl and LLVM 21.1.6 latest clang-cl)

CPython versions tested on:

3.14

Operating systems tested on:

Windows, Linux

[chris-eibl]

I've just verified that I can still successfully build 3.14.0 using Visual Studio 2026 Community Edition which bundles clang 20.1.8 (if you select it in the installer)

• build.bat --pgo --tail-call-interp "/p:PlatformToolset=ClangCL"
• build.bat -c Release --tail-call-interp "/p:PlatformToolset=ClangCL"
• build.bat -c Debug --tail-call-interp "/p:PlatformToolset=ClangCL"

which create x86_64 64bit tc binaries.

I've also just tested to build a 32bit binary

• build.bat -p Win32 -c Debug --tail-call-interp "/p:PlatformToolset=ClangCL"

Also the current tip of the 3.14 branch (64d6bde) builds fine using

• build.bat -c Release --tail-call-interp "/p:PlatformToolset=ClangCL"

and without tc

• build.bat -c Release "/p:PlatformToolset=ClangCL"

I also did a lot of such builds during 3.14 development using VS 2022 and various different clang versions I have installed from https://github.com/llvm/llvm-project/releases and passing the additional "/p:LLVMInstallDir=<my-clang-dir>" and "/p:LLVMToolsVersion=<my-clang-major-version>" parameters, e.g.

• build.bat --pgo "/p:PlatformToolset=ClangCL" "/p:LLVMInstallDir=C:\clang+llvm-20.1.6-x86_64-pc-windows-msvc" "/p:LLVMToolsVersion=20"

See more in section Building Python using Clang/LLVM in PCBuild/readme.txt.

Latest main builds for me as well, which I do on a regular basis.

How do you build? Do above commands work for you?

[winapiadmin]

tail-call only works, combining with JIT fails (?), combining with JIT tier 2 fails with this issue.

[winapiadmin]

Weirdly.... tail-call and JIT (additionally tier 2) are seemingly exclusive when I test. So we can't get full potential speed of tail calling and JIT interpreter.

[winapiadmin]

Also happens on linux?

[winapiadmin@windows cpython]$ make
clang -c -fno-strict-overflow -Wsign-compare -Wunreachable-code -DNDEBUG -g -O3 -Wall    -std=c11 -Wextra -Wno-unused-parameter -Wno-missing-field-initializers -Wstrict-prototypes -Werror=implicit-function-declaration -fvisibility=hidden -D_Py_TIER2=4  -I./Include/internal -I./Include/internal/mimalloc  -I. -I./Include    -DPy_BUILD_CORE -o Python/ceval.o Python/ceval.c
In file included from Python/ceval.c:1094:
Python/generated_cases.c.h:5584:13: error: use of undeclared identifier 'next_uop'
 5584 |             GOTO_TIER_TWO(executor);
      |             ^~~~~~~~~~~~~~~~~~~~~~~
Python/ceval_macros.h:392:5: note: expanded from macro 'GOTO_TIER_TWO'
  392 |     next_uop = _executor->trace; \
      |     ^~~~~~~~
In file included from Python/ceval.c:1094:
Python/generated_cases.c.h:5584:13: error: use of undeclared label 'enter_tier_two'
Python/ceval_macros.h:394:10: note: expanded from macro 'GOTO_TIER_TWO'
  394 |     goto enter_tier_two; \
      |          ^
In file included from Python/ceval.c:1094:
Python/generated_cases.c.h:7842:25: error: use of undeclared identifier 'next_uop'
 7842 |                         GOTO_TIER_TWO(executor);
      |                         ^~~~~~~~~~~~~~~~~~~~~~~
Python/ceval_macros.h:392:5: note: expanded from macro 'GOTO_TIER_TWO'
  392 |     next_uop = _executor->trace; \
      |     ^~~~~~~~
In file included from Python/ceval.c:1094:
Python/generated_cases.c.h:7842:25: error: use of undeclared label 'enter_tier_two'
Python/ceval_macros.h:394:10: note: expanded from macro 'GOTO_TIER_TWO'
  394 |     goto enter_tier_two; \
      |          ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:30:13: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   30 |             oparg = CURRENT_OPARG();
      |             ^~~~~
      |             optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:30:19: error: incompatible integer to pointer conversion assigning to 'char *' from 'const uint16_t' (aka 'const unsigned short') [-Wint-conversion]
   30 |             oparg = CURRENT_OPARG();
      |                   ^ ~~~~~~~~~~~~~~~
Python/executor_cases.c.h:31:18: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   31 |             if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) {
      |                  ^~~~~
      |                  optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:31:24: error: invalid operands to binary expression ('char *' and 'int')
   31 |             if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) {
      |                  ~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~
Python/executor_cases.c.h:79:13: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   79 |             oparg = CURRENT_OPARG();
      |             ^~~~~
      |             optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:79:19: error: incompatible integer to pointer conversion assigning to 'char *' from 'const uint16_t' (aka 'const unsigned short') [-Wint-conversion]
   79 |             oparg = CURRENT_OPARG();
      |                   ^ ~~~~~~~~~~~~~~~
Python/executor_cases.c.h:80:44: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   80 |             _PyStackRef value_s = GETLOCAL(oparg);
      |                                            ^~~~~
      |                                            optarg
Python/ceval_macros.h:230:44: note: expanded from macro 'GETLOCAL'
  230 | #define GETLOCAL(i)     (frame->localsplus[i])
      |                                            ^
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:80:35: error: array subscript is not an integer
   80 |             _PyStackRef value_s = GETLOCAL(oparg);
      |                                   ^        ~~~~~
Python/ceval_macros.h:230:43: note: expanded from macro 'GETLOCAL'
  230 | #define GETLOCAL(i)     (frame->localsplus[i])
      |                                           ^~
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:85:82: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   85 |                     PyTuple_GetItem(_PyFrame_GetCode(frame)->co_localsplusnames, oparg)
      |                                                                                  ^~~~~
      |                                                                                  optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:85:82: error: incompatible pointer to integer conversion passing 'char *' to parameter of type 'Py_ssize_t' (aka 'long') [-Wint-conversion]
   85 |                     PyTuple_GetItem(_PyFrame_GetCode(frame)->co_localsplusnames, oparg)
      |                                                                                  ^~~~~
./Include/tupleobject.h:32:62: note: passing argument to parameter here
   32 | PyAPI_FUNC(PyObject *) PyTuple_GetItem(PyObject *, Py_ssize_t);
      |                                                              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:99:13: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
   99 |             oparg = 0;
      |             ^~~~~
      |             optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:102:45: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
  102 |             value = PyStackRef_DUP(GETLOCAL(oparg));
      |                                             ^~~~~
      |                                             optarg
Python/ceval_macros.h:230:44: note: expanded from macro 'GETLOCAL'
  230 | #define GETLOCAL(i)     (frame->localsplus[i])
      |                                            ^
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:102:36: error: array subscript is not an integer
  102 |             value = PyStackRef_DUP(GETLOCAL(oparg));
      |                                    ^        ~~~~~
Python/ceval_macros.h:230:43: note: expanded from macro 'GETLOCAL'
  230 | #define GETLOCAL(i)     (frame->localsplus[i])
      |                                           ^~
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:111:13: error: use of undeclared identifier 'oparg'; did you mean 'optarg'?
  111 |             oparg = 1;
      |             ^~~~~
      |             optarg
/usr/include/bits/getopt_core.h:36:14: note: 'optarg' declared here
   36 | extern char *optarg;
      |              ^
In file included from Python/ceval.c:1291:
Python/executor_cases.c.h:111:19: error: incompatible integer to pointer conversion assigning to 'char *' from 'int' [-Wint-conversion]
  111 |             oparg = 1;
      |                   ^ ~
fatal error: too many errors emitted, stopping now [-ferror-limit=]
20 errors generated.
make: *** [Makefile:3297: Python/ceval.o] Error 1

Configure command: ./configure --with-tail-call-interp --enable-experimental-jit=interpreter CXX=clang++ CC=clang

This also fails with GCC. with more warnings. (JIT requires LLVM)

[chris-eibl]

tail-call only works, combining with JIT fails (?), combining with JIT tier 2 fails with this issue.

64d6bde (now only almost current tip of the 3.14 branch :) builds fine for me on Windows with tc and jit using

build.bat --experimental-jit --tail-call-interp "/p:PlatformToolset=ClangCL" "/p:LLVMInstallDir=<your-repo-root>\externals\llvm-19.1.7.0" "/p:LLVMToolsVersion=19"

I highly recommend pinning clang to the same version as used for the jit stencils (<your-repo-root>\externals\llvm-19.1.7.0 in case of 3.14), because e.g.

"/p:LLVMInstallDir=<my-path-to>\llvm-21.1.4.0"

smuggles its include path in front of the environment variable

INCLUDE=<my-path-to>\llvm-21.1.4.0\lib\clang\21\include;<other include paths>

and then building of the jit stencils can fail (but that's totally different errors than those you mentioned above) ¹.

Footnotes

1. I think the clang version used for building the stencils on Linux is not so tightly coupled to the one used to compile Python itself.
   Furthermore, since https://github.com/python/cpython/issues/138497 it is possible to switch the clang version used for the stencil generation, but that's an environment variable used during configure which then sets a new command line option when invoking the generation of the stencils (--llvm-version=\"$LLVM_VERSION\"), so that's no help on Windows ... ↩

[chris-eibl]

Weirdly.... tail-call and JIT (additionally tier 2) are seemingly exclusive when I test. So we can't get full potential speed of tail calling and JIT interpreter.

Ah, it seems you are using --experimental-jit-interpreter, AKA "tier 2".
This one fails for me on 3.14 on Windows with the errors you've listed, too, but is working on latest main.

[chris-eibl]

Configure command: ./configure --with-tail-call-interp --enable-experimental-jit=interpreter CXX=clang++ CC=clang

Ah, again --enable-experimental-jit=interpreter: you are not using the jit when specifying this, but rather the tier 2 interpreter,
which is mostly useful for debugging purposes (for JIT developers).

See https://docs.python.org/dev/using/configure.html#cmdoption-enable-experimental-jit.

Latest main should work on Linux as well, @Fidget-Spinner did a lot of builds during his work on #139109 which at the beginning even mandated tail calling (since #141703 the tracing jit supports non-tc interpreters again).