gh-124552 : Improve the accuracy of possible breakpoint check in bdb #124553
Conversation
gaogaotiantian
changed the title
|
This needs a test. |
Lib/bdb.py
Outdated
| self.code_lineno[code] = set() | ||
| for _, _, lineno in code.co_lines(): | ||
| self.code_lineno[code].add(lineno) |
There was a problem hiding this comment.
| self.code_lineno[code] = set() | |
| for _, _, lineno in code.co_lines(): | |
| self.code_lineno[code].add(lineno) | |
| self.code_lineno[code] = set([lno for _, _, lno in code.co_lines()]) |
|
Do you think a test that directly tests |
|
I added a test for |
|
I am testing this with IDLE's bdb-based debugger now. |
|
No (new) problems found. Breakpoints at top level and in def still work. |
|
Thanks @terryjreedy for confirming this. #124533 might be another interesting one to test. |
Lib/bdb.py
Outdated
| return False | ||
| if code not in self.code_lineno: | ||
| self.code_lineno[code] = set(lineno for _, _, lineno in code.co_lines()) | ||
| return lineno in self.code_lineno[frame.f_code] |
There was a problem hiding this comment.
| return lineno in self.code_lineno[frame.f_code] | |
| return lineno in self.code_lineno[code] |
Lib/bdb.py
Outdated
| @@ -36,6 +37,7 @@ def __init__(self, skip=None): | |||
| self.frame_returning = None | |||
| self.trace_opcodes = False | |||
| self.enterframe = None | |||
| self.code_lineno = weakref.WeakKeyDictionary() | |||
There was a problem hiding this comment.
| self.code_lineno = weakref.WeakKeyDictionary() | |
| self.code_linenos = weakref.WeakKeyDictionary() |
Misc/NEWS.d/next/Library/2024-09-25-22-06-52.gh-issue-124552.1nQKNM.rst
Outdated
Show resolved
Hide resolved
Lib/bdb.py
Outdated
| @@ -275,7 +280,23 @@ def do_clear(self, arg): | |||
| def break_anywhere(self, frame): | |||
| """Return True if there is any breakpoint for frame's filename. | |||
There was a problem hiding this comment.
this comment needs updating (it's not just filename now).
…nQKNM.rst Co-authored-by: Irit Katriel <1055913+iritkatriel@users.noreply.github.com>
| @@ -155,6 +157,9 @@ def dispatch_return(self, frame, arg): | |||
| if self.stop_here(frame) or frame == self.returnframe: | |||
| # Ignore return events in generator except when stepping. | |||
| if self.stopframe and frame.f_code.co_flags & GENERATOR_AND_COROUTINE_FLAGS: | |||
| # It's possible to trigger a StopIteration exception in | |||
| # the caller so we must set the trace function in the caller | |||
| self._set_caller_tracefunc(frame) | |||
There was a problem hiding this comment.
how is this related to this PR?
There was a problem hiding this comment.
There is a hidden bug which was exposed by the change of break_anywhere. There's a test case of bdb testing raising StopIteration in a generator. It passed because break_anywhere always return True when the function was defined in the same file, which always sets the trace function on the caller because it is in the same file. That coincidence hide the bug where returning from a generator should stop in the caller. There's a similar call in line 177 - that's for the normal return case from #118979 - it's also my fix to a similar case.
Now we check whether the line number is in the code object with
co_lines()so we can have an accurate answer.A bug needs to be fixed for generators. This is not caused by the new feature, but exposed. I put those together because the changes are very small. I can separate those if that's desired.