On Nov 9, 2017, at 2:44 AM, Serhiy Storchaka wrote:

If the problem is with naming, what names do you prefer? This already was bikeshedded (I insisted on discussing names before introducing the macros), but may now you have better ideas?

It didn't really seem like a bad idea until after you swept through the code with 200+ applications of the macro and I saw how unclear the results were. Even code that I wrote myself is now harder for me to grok (for example, the macro was applied 17 times to already correct code in itertools). We used to employ a somewhat plain coding style that was easy to walk through, but the following examples seem opaque. I find it takes practice to look at any one of these and say that it is unequivocally correct (were the function error return arguments handled correctly, are the typecasts proper, at what point can a reentrant call occur, which is the source operand and which is the destination, is the macro using either of the operands twice, is the destination operand an allowable lvalue, do I need to decref the source operand afterwards, etc): Py_SETREF(((PyHeapTypeObject*)type)->ht_name, value) Py_SETREF(newconst, PyFrozenSet_New(newconst)); Py_XSETREF(c->u->u_private, s->v.ClassDef.name); Py_SETREF(*p, t); Py_XSETREF(self->lineno, PyTuple_GET_ITEM(info, 1)); Py_SETREF(entry->path, PyUnicode_EncodeFSDefault(entry->path)); Py_XSETREF(self->checker, PyObject_GetAttrString(ob, "_check_retval_")); Py_XSETREF(fut->fut_source_tb, _PyObject_CallNoArg(traceback_extract_stack)); Stylistically, all of these seem awkward and I think there is more to it than just the name. I'm not sure it is wise to pass complex inputs into a two-argument macro that makes an assignment and has a conditional refcount side-effect. Even now, one of the above looks to me like it might not be correct. Probably, we're the wrong people to be talking about this. The proposal is to make these macros part of the official API so that it starts to appear in source code everywhere. The question isn't whether the above makes sense to you and me; instead, it is whether other people can make heads or tails out the above examples. As a result of making the macros official, will the Python world have a net increase in complexity or decrease in complexity? My personal experience with the macros hasn't been positive. Perhaps everyone else thinks it's fine. If so, I won't stand in your way. Raymond

On 9 November 2017 at 22:35, Antoine Pitrou wrote:

On Thu, 9 Nov 2017 04:22:20 -0800 Raymond Hettinger wrote:

...

Probably, we're the wrong people to be talking about this. The proposal is to make these macros part of the official API so that it starts to appear in source code everywhere. The question isn't whether the above makes sense to you and me; instead, it is whether other people can make heads or tails out the above examples.

Generally I would advocate anyone wanting to write a third-party C extension, but not very familiar with the C API and its quirks, use Cython instead. I'm not sure if that's an argument for the SETREF APIs to remain private or to become public :-)

I'm with Antoine on this - we should be pushing folks writing extension modules towards code generators like Cython, cffi, SWIG, and SIP, support libraries like Boost::Python, or safer languages like Rust (which can then be wrapped with cffi), rather than encouraging more bespoke C/C++ extensions modules with handcrafted refcount management. There's a reason the only parts of https://packaging.python.org/guides/packaging-binary-extensions/ that have actually been filled in are the ones explaining how to use a tool to write the extension module for you :) For me, that translates to being -1 on making these part of the public API - for code outside CPython, our message should consistently be "Do not roll your own refcount management, get a code generator or library to handle it for you". Cheers, Nick. -- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia

Hum, to give more context to the discussion, the two discussed macros are documented this way: #ifndef Py_LIMITED_API /* Safely decref `op` and set `op` to `op2`. * * As in case of Py_CLEAR "the obvious" code can be deadly: * * Py_DECREF(op); * op = op2; * * The safe way is: * * Py_SETREF(op, op2); * * That arranges to set `op` to `op2` _before_ decref'ing, so that any code * triggered as a side-effect of `op` getting torn down no longer believes * `op` points to a valid object. * * Py_XSETREF is a variant of Py_SETREF that uses Py_XDECREF instead of * Py_DECREF. */ #define Py_SETREF(op, op2) \ do { \ PyObject *_py_tmp = (PyObject *)(op); \ (op) = (op2); \ Py_DECREF(_py_tmp); \ } while (0) #define Py_XSETREF(op, op2) \ do { \ PyObject *_py_tmp = (PyObject *)(op); \ (op) = (op2); \ Py_XDECREF(_py_tmp); \ } while (0) #endif /* ifndef Py_LIMITED_API */ Victor 2017-11-09 13:22 GMT+01:00 Raymond Hettinger :

...

On Nov 9, 2017, at 2:44 AM, Serhiy Storchaka wrote:

If the problem is with naming, what names do you prefer? This already was bikeshedded (I insisted on discussing names before introducing the macros), but may now you have better ideas?

It didn't really seem like a bad idea until after you swept through the code with 200+ applications of the macro and I saw how unclear the results were. Even code that I wrote myself is now harder for me to grok (for example, the macro was applied 17 times to already correct code in itertools).

We used to employ a somewhat plain coding style that was easy to walk through, but the following examples seem opaque. I find it takes practice to look at any one of these and say that it is unequivocally correct (were the function error return arguments handled correctly, are the typecasts proper, at what point can a reentrant call occur, which is the source operand and which is the destination, is the macro using either of the operands twice, is the destination operand an allowable lvalue, do I need to decref the source operand afterwards, etc):

Py_SETREF(((PyHeapTypeObject*)type)->ht_name, value) Py_SETREF(newconst, PyFrozenSet_New(newconst)); Py_XSETREF(c->u->u_private, s->v.ClassDef.name); Py_SETREF(*p, t); Py_XSETREF(self->lineno, PyTuple_GET_ITEM(info, 1)); Py_SETREF(entry->path, PyUnicode_EncodeFSDefault(entry->path)); Py_XSETREF(self->checker, PyObject_GetAttrString(ob, "_check_retval_")); Py_XSETREF(fut->fut_source_tb, _PyObject_CallNoArg(traceback_extract_stack));

Stylistically, all of these seem awkward and I think there is more to it than just the name. I'm not sure it is wise to pass complex inputs into a two-argument macro that makes an assignment and has a conditional refcount side-effect. Even now, one of the above looks to me like it might not be correct.

Probably, we're the wrong people to be talking about this. The proposal is to make these macros part of the official API so that it starts to appear in source code everywhere. The question isn't whether the above makes sense to you and me; instead, it is whether other people can make heads or tails out the above examples. As a result of making the macros official, will the Python world have a net increase in complexity or decrease in complexity?

My personal experience with the macros hasn't been positive. Perhaps everyone else thinks it's fine. If so, I won't stand in your way.

Raymond

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/victor.stinner%40gmail.co...

Recently, Oren Milman fixed multiple bugs when an __init__() method was called twice. IMHO Py_SETREF() was nicely used in __init__(): https://github.com/python/cpython/commit/e56ab746a965277ffcc4396d8a0902b6e07... https://github.com/python/cpython/commit/c0cabc23bbe474d542ff8a4f1243f4ec3cc... While it's possible to rewrite the code *correctly* without PY_SETREF(), it would be much more verbose. Here the fix remains a single line: - self->archive = filename; + Py_XSETREF(self->archive, filename); Victor 2017-11-09 3:08 GMT+01:00 Raymond Hettinger :

...

On Nov 8, 2017, at 8:30 AM, Serhiy Storchaka wrote:

Macros Py_SETREF and Py_XSETREF were introduced in 3.6 and backported to all maintained versions ([1] and [2]). Despite their names they are private. I think that they are enough stable now and would be helpful in third-party code. Are there any objections against adding them to the stable C API? [3]

I have mixed feeling about this. You and Victor seem to really like these macros, but they have been problematic for me. I'm not sure whether it is a conceptual issue or a naming issue, but the presence of these macros impairs my ability to read code and determine whether the refcounts are correct. I usually end-up replacing the code with the unrolled macro so that I can count the refs across all the code paths.

The other issue is that when there are multiple occurrences of these macros for multiple variables, it interferes with my best practice of deferring all decrefs until the data structures are in a fully consistent state. Any one of these can cause arbitrary code to run. I greatly prefer putting all the decrefs at the end to increase my confidence that it is okay to run other code that might reenter the current code. Pure python functions effectively have this built-in because the locals all get decreffed at the end of the function when a return-statement is encountered. That practice helps me avoid hard to spot re-entrancy issues.

Lastly, I think we should have a preference to not grow the stable C API. Bigger APIs are harder to learn and remember, not so much for you and Victor who use these frequently, but for everyone else who has to lookup all the macros whose function isn't immediately self-evident.

Raymond

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/victor.stinner%40gmail.co...