f_trace_opcodes frame attribute to switch to per-opcode tracing

[ncoghlan]

BPO	31344
Nosy	@gpshead, @ncoghlan, @nedbat, @njsmith, @serhiy-storchaka, @zooba
PRs	bpo-31344: Per-frame control of trace events #3417

^{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 = 'https://github.com/ncoghlan'
closed_at = <Date 2017-09-08.00:14:59.760>
created_at = <Date 2017-09-04.22:03:40.531>
labels = ['type-feature', '3.7']
title = 'f_trace_opcodes frame attribute to switch to per-opcode tracing'
updated_at = <Date 2017-09-08.00:14:59.759>
user = 'https://github.com/ncoghlan'

bugs.python.org fields:

activity = <Date 2017-09-08.00:14:59.759>
actor = 'ncoghlan'
assignee = 'ncoghlan'
closed = True
closed_date = <Date 2017-09-08.00:14:59.760>
closer = 'ncoghlan'
components = []
creation = <Date 2017-09-04.22:03:40.531>
creator = 'ncoghlan'
dependencies = []
files = []
hgrepos = []
issue_num = 31344
keywords = []
message_count = 6.0
messages = ['301273', '301362', '301366', '301374', '301565', '301655']
nosy_count = 6.0
nosy_names = ['gregory.p.smith', 'ncoghlan', 'nedbat', 'njs', 'serhiy.storchaka', 'steve.dower']
pr_nums = ['3417']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue31344'
versions = ['Python 3.7']

[ncoghlan]

In order to test bpo-29988 reliably, we want the ability to run trace hooks for every opcode in a frame, rather than for every line.

The simplest API we've been able to come up with for that is a "f_trace_opcodes" attribute on the frame which we set from the trace hook the first time that it runs.

[serhiy-storchaka]

The f_trace attribute is writable from a tracing function. There are different restrictions, for example you can't jump inside a 'for' loop from outside. The code of the setter is complex and hard for modifications (see the discussion on PR 2827). Allowing to trace opcodes can make it more complex.

[njsmith]

Adding Ned to CC in case he wants to comment on the utility of per-opcode tracing from the perspective of coverage.py.

[ncoghlan]

Steve's also requested the ability to *turn off* line debugging, and I'm thinking it may make sense to allow all four configurations:

• both line and opcode events disabled
• only getting one or the other
• getting both kinds of event

Opcode events would definitely be off by default, and if you do turn them on, then you'd be in a situation similar to importing ctypes: the interpreter is no longer guaranteed to work correctly. (Although unlike ctypes, you shouldn't get segfaults because of it)

[ncoghlan]

My posted PR implements the "independent flags" option, where f_trace_lines and f_trace_opcodes allow per-line and per-opcode events to be requested independently.

The default is to emit per-line events (as has historically been the case).

[ncoghlan]

New changeset 5a85167 by Nick Coghlan in branch 'master':
bpo-31344: Per-frame control of trace events (GH-3417)
5a85167