python: Fix GC isn't working correctly. #1888
Conversation
There was a problem hiding this comment.
Thanks for working on this! But I'm afraid it won't work well because get_new_sym_addr() always returns a new address even for the same name.
I think you need to compare func_name instead. It'd have little more overhead but probably better to make GC working properly.
python/trace-python.c
Outdated
| /* just compare pointers of the code object */ | ||
| if (iter->code == code) { | ||
| Py_DECREF(code); | ||
| if (iter->addr == new_addr) { |
yihong0618
force-pushed
the
fix_issue_1886
branch
4 times, most recently
from
1e0f0ba
to
efbe17c
Compare
|
Fixed thanks for the review, And use func_name is indeed a better way. and uftrace_py_traverse and uftrace_py_clean is no need either. |
python/trace-python.c
Outdated
| if (iter->code == code) { | ||
| Py_DECREF(code); | ||
| /* compare func_name */ | ||
| if (iter->name == func_name) { |
There was a problem hiding this comment.
I don't think we can do this. It should call strcmp().
yihong0618
force-pushed
the
fix_issue_1886
branch
from
efbe17c
to
4846f76
Compare
The uftrace_python_symbol keeps code PyObject in rb tree so GC can not run correctly in the trace process. Fixed the problem by dropping it. and change rb compare from code PyObject to func_name Fixed: namhyung#1886 Signed-off-by: Yi Hong <zouzou0208@gmail.com> Signed-off-by: Honggyu Kim <honggyu.kp@gmail.com>
yihong0618
force-pushed
the
fix_issue_1886
branch
from
4846f76
to
25e93b9
Compare
|
Hi @yihong0618, thanks very much for the fix. I've also found that it fixes the problem I reported. But you don't have to put my |
learned. will remember that next thanks a lot^_^ |
The uftrace_python_symbol keeps code PyObject in rb tree so GC can not run correctly in the trace process. Fixed the problem by dropping it.
Fixed: #1886