| msg72127 - (view) |
Author: Amaury Forgeot d'Arc (amaury.forgeotdarc) *  |
Date: 2008-08-29 12:11 |
| The example Demo/embed/importexc.c crashes, because Py_NewInterpreter cannot reimport builtins and sys modules. This problem seems important for embedding applications like mod_python, for example. (the "import exceptions" statement does not work with python 3.0, but replacing with e.g. "import types" does not change anything: the interpreter is not correctly renewed) I tried to put "-1" in the m_size structure of these modules, and they seem to import correctly. However, "builtins" comes with its original dictionary - without the standard exceptions. I think that these modules should be made re-importable, with specific functions. Maybe two related problems: - once you've done del sys.modules['sys'] it's not possible to get it back, "import sys" raises an error... - the usual trick to call sys.setdefaultencoding will not work, since it needs to "imp.reload(sys)" |
|
|
| msg73403 - (view) |
Author: Benjamin Peterson (benjamin.peterson) * |
Date: 2008-09-18 22:15 |
| Maybe, I'm not seeing the whole problem, but can't we just add _PySys_Init and _PyBuiltin_Init to config.c like in the attached patch? Obviously, we will eventually want to make a separate state to store module globals in, but I think this will work for 3.0 final. |
|
|
| msg73530 - (view) |
Author: Amaury Forgeot d'Arc (amaury.forgeotdarc) * |
Date: 2008-09-21 21:02 |
| I applied the patch to PC/config.c, but this did not change anything. |
|
|
| msg73531 - (view) |
Author: Benjamin Peterson (benjamin.peterson) * |
Date: 2008-09-21 21:15 |
| Interesting, here it lets import.c's init_builtin reinitalize modules... |
|
|
| msg74069 - (view) |
Author: Graham Dumpleton (grahamd) |
Date: 2008-09-30 12:15 |
| Adding the functions as initfunc in module init table is of no use as they aren't invoked when creating a sub interpreter. One thing that does appear to work, although no idea of whether it is correct way to solve problem, is to duplicate the builtin/sys initialisation that occurs in Py_InitializeEx() function. Attached diff shows nature of changes. Diff is bit messy as have left existing code in there but #ifdef'd out. Maybe this will give someone who knows how overall interpreter initialisation is supposed to work a head start on coming up with proper fix. But then it could be totally wrong as well. At least with change as is, mod_wsgi works for sub interpreters now. I'll do more work later on whether it is correct way to solve it. |
|
|
| msg74070 - (view) |
Author: Amaury Forgeot d'Arc (amaury.forgeotdarc) * |
Date: 2008-09-30 12:22 |
| Your patch may go in the right direction, but please provide only context diff or unified diff files. Use "diff -du", or "svn diff" to generate the file. |
|
|
| msg74071 - (view) |
Author: Graham Dumpleton (grahamd) |
Date: 2008-09-30 12:26 |
| Argh. Personally I like to provide context diff's but more often than not get abused for providing them over a unified diff. Was in a hurry this time as had only a couple of minutes of battery life left on the laptop, so quickly did it without thinking and then ran off to find a power point. :-) |
|
|
| msg74072 - (view) |
Author: Graham Dumpleton (grahamd) |
Date: 2008-09-30 12:31 |
| Unified diff now attached. |
|
|
| msg74919 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2008-10-17 15:47 |
| Here is a patch that fixed importexc.c. It consists of the following parts: - set m_size of the builtins module and the sys module to -1, indicating that these modules don't support repeated initialization. This should be reviewed; perhaps it's better (and necessary) to record the init function not only for dynamically-loaded modules in m_init, but also for statically linked ones, so that the reinit code can call them again (whether or not it is safe to call them again for sys and builtins should then be studied). - add a field to the interpreter state indicating that the codecs are not ready. when trying to use the file system encoding in this state, use ASCII instead. - Fix importexc to use the types module. |
|
|
| msg74938 - (view) |
Author: Amaury Forgeot d'Arc (amaury.forgeotdarc) * |
Date: 2008-10-17 21:24 |
| I think the patch goes in the right direction. But in addition, Py_NewInterpreter() has to call initstdio() between initmain() and initsite() (the same sequence as in Py_InitializeEx) Found by using the following command string in importexc.c: "import types; print(types.XXX)" |
|
|
| msg74939 - (view) |
Author: Benjamin Peterson (benjamin.peterson) * |
Date: 2008-10-17 21:27 |
| Wouldn't it make more sense to move interpreter initialization things to Py_NewInterpreter and call it from Py_InitializeEx? |
|
|
| msg74940 - (view) |
Author: Christian Heimes (christian.heimes) * |
Date: 2008-10-17 21:29 |
| Sounds like a good plan, Benjamin |
|
|
| msg74943 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2008-10-17 22:34 |
| > Wouldn't it make more sense to move interpreter initialization things to > Py_NewInterpreter and call it from Py_InitializeEx? Can you propose a specific patch? I'm worried that doing so blindly introduces other bugs. |
|
|
| msg75248 - (view) |
Author: Christian Heimes (christian.heimes) * |
Date: 2008-10-26 22:58 |
| The patch "subinterpreter.patch" is based on Martin's patch "importexc.diff". The patch contains additional code to setup a preliminary stderr object and a call to initstdio(). Amaury is right. I had to add initstdio() to initialize the standard streams. But I can't get it to work. $ rm -f Demo/embed/importexc.o; cd Demo/embed; make; ./importexc; cd ../.. gcc -g -I../../Include -I../.. -c -o importexc.o importexc.c gcc -Xlinker -export-dynamic importexc.o ../../libpython3.0.a -lnsl -ldl -lreadline -ltermcap -lieee -lpthread -lutil -lm -o importexc Initialize interpreter <module 'types' from '/usr/local/lib/python3.0/types.py'> Initialize subinterpreter Fatal Python error: Py_Initialize: can't initialize sys standard streams Traceback (most recent call last): File "/usr/local/lib/python3.0/encodings/__init__.py", line 32, in ValueError: Cannot encode path item Aborted |
|
|
| msg75277 - (view) |
Author: Christian Heimes (christian.heimes) * |
Date: 2008-10-28 11:20 |
| In combination with the patch in #4213, "subinterpreter.patch" fixes the problem. I'm assigning the bug to Barry for his final decision. |
|
|
| msg75338 - (view) |
Author: Graham Dumpleton (grahamd) |
Date: 2008-10-30 00:10 |
| In conjunction with #4213, the attached subinterpreter.patch appears to fix issue for mod_wsgi. |
|
|
| msg75385 - (view) |
Author: Christian Heimes (christian.heimes) * |
Date: 2008-10-30 21:48 |
| Applied in r67057 |
|
|