[pypy-dev] cannot use "id" in Rpython
Shubha Ramani
shubharamani at yahoo.com
Fri Dec 30 13:10:49 EST 2016
I don't know the signature of the function coming into execute_frame ahead of time. Therefore as you suggested Manuel,I am leaving the argument is an empty list []. This doesn't seem to work though. See the assertion error I'm getting below.All other examples of this technique I see being used are passing in specific argument types to define a succinctfunction signature, i.e. in test_warmstate.py: ENTER_JIT = lltype.Ptr(lltype.FuncType([lltype.Signed, lltype.Float, lltype.Signed], lltype.Bool))
The following causes an assertion error:
fun_name = _safe(pycode.co_name) fptr_type = lltype.Ptr(lltype.FuncType([], lltype.Void)) fptr = llhelper(fptr_type, fun_name)
[translation] start debugger...> /opt/shubha_vtune_pypy/rpython/rtyper/lltypesystem/lltype.py(776)constPtr()-> assert T.is_constant()
On Friday, December 30, 2016 8:59 AM, Shubha Ramani <shubharamani at yahoo.com> wrote:
This is very helpful. Thank you Manuel !
I will try your suggestion.
Shubha
> On Dec 30, 2016, at 8:04 AM, Manuel Jacob <me at manueljacob.de> wrote:
>
> The id of an object is not necessarily related in any way to the address of an object in Python.
>
> In RPython, you can convert a function to a low-level function pointer with llhelper() from rpython.rtyper.annlowlevel. You can call it like this:
>
> fptr_type = lltype.Ptr(lltype.FuncType([], lltype.Void))
> fptr = llhelper(fptr_type, func)
>
> Depending on what you want to do with the function address exactly, you might need to cast the function pointer to a RPython "Address" with cast_ptr_to_adr() from rpython.rtyper.lltypesystem.llmemory.
>
>> On 2016-12-30 14:42, Shubha Ramani via pypy-dev wrote:
>> It's not a bad idea for me. I'm hooking up to a tool made by intel
>> called "vtune" which requires the function name, function size and
>> function address from the original python script code. I got the first
>> two from Python reflection (the way VmProf does it) but I need the
>> third (original script function address). Is __repr__ supported in
>> Rpython ? The string returned by __repr__ contains the Python address
>> at least in Python. If id() doesn't work but __repr__ does then I can
>> use it in the overloaded execute_frame and that should be fine.
>> Shubha
>>> On Dec 29, 2016, at 4:29 PM, Ryan Gonzalez <rymg19 at gmail.com> wrote:
>>> What exactly are you trying to do. Using id to get the address of a function is usually a bad idea...
>>> --
>>> Ryan (ライアン)
>>> Yoko Shimomura > ryo (supercell/EGOIST) > Hiroyuki Sawano >> everyone else
>>> http://kirbyfan64.github.io/
>>>> On Dec 29, 2016 1:19 PM, "Shubha Ramani via pypy-dev" <pypy-dev at python.org> wrote:
>>>> What can I use then ? I need the address of the function ?
>>>> [translation:ERROR] Exception: cannot use id() in RPython; see objectmodel.compute_xxx()
>>>> Processing block:
>>>> block at 19 is a <class 'rpython.flowspace.flowcontext.SpamBlock'>
>>>> in (pypy.module._vtune.interp_vtune:28)_get_full_name
>>>> containing the following operations:
>>>> buf_0 = simple_call((type RStringIO))
>>>> v943 = getattr(buf_0, ('write'))
>>>> v944 = getattr(pycode_0, ('co_code'))
>>>> v945 = simple_call(v943, v944)
>>>> v946 = getattr(buf_0, ('getsize'))
>>>> size_code_0 = simple_call(v946)
>>>> v947 = getattr(pycode_0, ('co_name'))
>>>> id_func_0 = id(v947)
>>>> v948 = getattr(pycode_0, ('co_name'))
>>>> v949 = simple_call((function _safe), v948)
>>>> v950 = getattr(pycode_0, ('co_firstlineno'))
>>>> v951 = getattr(pycode_0, ('co_filename'))
>>>> v952 = simple_call((function _safe), v951)
>>>> v953 = newtuple(v949, v950, v952, size_code_0, id_func_0)
>>>> v954 = mod(('vtune:%s:%d:%s:%d:%d'), v953)
>>>> --end--
>>>> [translation] start debugger...
>>>> > /opt/shubha_vtune_pypy/rpython/annotator/unaryop.py(188)id()
>>>> -> raise Exception("cannot use id() in RPython; "
>>>> (Pdb+)
>>>> _______________________________________________
>>>> pypy-dev mailing list
>>>> pypy-dev at python.org
>>>> https://mail.python.org/mailman/listinfo/pypy-dev
>> _______________________________________________
>> pypy-dev mailing list
>> pypy-dev at python.org
>> https://mail.python.org/mailman/listinfo/pypy-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/pypy-dev/attachments/20161230/62d4af65/attachment-0003.html>
More information about the pypy-dev
mailing list