Add f_state attribute to FrameObjects.

[markshannon]

BPO	47046
Nosy	@nedbat, @markshannon, @brandtbucher

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2022-03-17.16:40:39.177>
labels = []
title = 'Add `f_state` attribute to FrameObjects.'
updated_at = <Date 2022-04-08.11:36:06.350>
user = 'https://github.com/markshannon'

bugs.python.org fields:

activity = <Date 2022-04-08.11:36:06.350>
actor = 'Mark.Shannon'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = []
creation = <Date 2022-03-17.16:40:39.177>
creator = 'Mark.Shannon'
dependencies = []
files = []
hgrepos = []
issue_num = 47046
keywords = []
message_count = 3.0
messages = ['415427', '415456', '416978']
nosy_count = 3.0
nosy_names = ['nedbat', 'Mark.Shannon', 'brandtbucher']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = None
url = 'https://bugs.python.org/issue47046'
versions = []

[markshannon]

When tracing, the event supplied is insufficient to determine what is actually happening.

E.g. A "call" event could be a call to a function or resuming a generator or coroutine.

Adding a state field to the FrameObject would allow these cases to be disambiguated without having to make dubious deductions from f_lasti or other frame attributes.

The proposed states would be:

FRAME_CREATED # Frame created, but not executed at all
FRAME_SUSPENDED # Frame suspended after yield or yield from
FRAME_EXECUTING # Frame is executed normally
FRAME_COMPLETED # Frame has completed
FRAME_CLEARED # Frame has been cleared

Ned, any other states that you might need to know about?

[nedbat]

These look great. Currently, coverage.py only tries to distinguish between return/yield, which I guess will now be COMPLETED and SUSPENDED?

[markshannon]

Don't you need to know if a "call" event is a call or the resumption of a generator?

[nedbat]

@markshannon Sorry, you are right, I also distinguish between call and resume:

#ifdef RESUME   // 3.11.0a4
    /*
     * The current opcode is guaranteed to be RESUME. The argument
     * determines what kind of resume it is.
     */
    PyObject * pCode = MyFrame_GetCode(frame)->co_code;
    real_call = (PyBytes_AS_STRING(pCode)[PyFrame_GetLasti(frame) + 1] == 0);
#else
    // f_lasti is -1 for a true call, and a real byte offset for a generator re-entry.
    real_call = (PyFrame_GetLasti(frame) < 0);
#endif

    if (real_call) {
        self->pcur_entry->last_line = -MyFrame_GetCode(frame)->co_firstlineno;
    }
    else {
        self->pcur_entry->last_line = PyFrame_GetLineNumber(frame);
    }

[markshannon]

As our automation doesn't seem to be quite there yet, here's a link to the PR implementing this:
#91443

[markshannon]

Unfortunately, this isn't useful for distinguishing events during tracing.

Because "call" events occur immediately after the call or resumption, then f_state is FRAME_EXECUTING for both calls and resumptions.
Similarly, "return" events occur immediately before the return or yield, so f_state is FRAME_EXECUTING is for both returns and yields.

We can't add new events, or change the sequencing without breaking things, so it looks like we will have to wait for PEP 669 and 3.12 to have a nice API for events. 😞

[nedbat]

Can you make a recommendation for the best way to make this distinction in 3.11?

[markshannon]

Can you make a recommendation for the best way to make this distinction in 3.11?

We aren't going to change anything for 3.11, so your check of the RESUME operand is probably best.

[nedbat]

I've fixed the coverage.py compilation errors in the nedbat/three-eleven-beta branch, but the results are wrong for generators. I'm not sure what detail I'm missing. @markshannon? @vstinner?

[vstinner]

I've fixed the coverage.py compilation errors

Which compiler errors? Which Python version are you using (release or the current main branch)?

[vstinner]

Is your comment related to nedbat/coveragepy#1367 (comment) issue?

[markshannon]

I think this is obsolete thanks to sys.monitoring.