* [PATCH 0/2] perf python: Add an example for sampling
@ 2025-07-28 5:59 Gautam Menghani
2025-07-28 5:59 ` [PATCH 1/2] perf python: Add support for record_opts struct Gautam Menghani
` (4 more replies)
0 siblings, 5 replies; 12+ messages in thread
From: Gautam Menghani @ 2025-07-28 5:59 UTC (permalink / raw)
To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin,
jolsa, irogers, adrian.hunter, kan.liang
Cc: Gautam Menghani, maddy, linux-perf-users, linux-kernel
Add an example for sampling events with python. Patch 1 adds support for
record_opts struct and patch 2 adds an example for sampling. The example
also demonstrates usage of the record_opts struct to customize the evsel
initialization from python.
Gautam Menghani (2):
perf python: Add support for record_opts struct
perf python: Add sampling.py as example for sampling
tools/perf/python/sampling.py | 49 +++++++++
tools/perf/util/python.c | 186 +++++++++++++++++++++++++++++++++-
2 files changed, 232 insertions(+), 3 deletions(-)
create mode 100755 tools/perf/python/sampling.py
--
2.49.0
^ permalink raw reply [flat|nested] 12+ messages in thread* [PATCH 1/2] perf python: Add support for record_opts struct 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani @ 2025-07-28 5:59 ` Gautam Menghani 2025-07-28 5:59 ` [PATCH 2/2] perf python: Add sampling.py as example for sampling Gautam Menghani ` (3 subsequent siblings) 4 siblings, 0 replies; 12+ messages in thread From: Gautam Menghani @ 2025-07-28 5:59 UTC (permalink / raw) To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang Cc: Gautam Menghani, maddy, linux-perf-users, linux-kernel Add support for record_opts struct to allow python to control the initialization of the evsels instead of using the current default options. Signed-off-by: Gautam Menghani <gautam@linux.ibm.com> --- tools/perf/util/python.c | 186 ++++++++++++++++++++++++++++++++++++++- 1 file changed, 183 insertions(+), 3 deletions(-) diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c index 2f28f71325a8..8537edbb3b0b 100644 --- a/tools/perf/util/python.c +++ b/tools/perf/util/python.c @@ -1045,6 +1045,169 @@ static int pyrf_evsel__setup_types(void) return PyType_Ready(&pyrf_evsel__type); } +struct pyrf_target { + PyObject_HEAD + + struct target target; +}; + +static int pyrf_target__init(struct pyrf_target *target, + PyObject *args, PyObject *kwargs) +{ + static char *kwlist[] = { "pid", "tid", "cpu_list", "bpf_str", "system_wide", "uses_mmap", + "default_per_cpu", "per_thread", "use_bpf", "inherit", + "initial_delay", "attr_map", NULL }; + + if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|ssssppppppis", + kwlist, &(target->target.pid), &(target->target.tid), + &(target->target.cpu_list), &(target->target.bpf_str), + &(target->target.system_wide), + &(target->target.uses_mmap), + &(target->target.default_per_cpu), + &(target->target.per_thread), &(target->target.use_bpf), + &(target->target.inherit), &(target->target.initial_delay), + &(target->target.attr_map))) + return -1; + + return 0; +} + +static void pyrf_target__delete(struct pyrf_target *target) +{ + Py_TYPE(target)->tp_free((PyObject *)target); +} + +static const char pyrf_target__doc[] = PyDoc_STR("target object."); + +static PyTypeObject pyrf_target__type = { + PyVarObject_HEAD_INIT(NULL, 0) + .tp_name = "perf.target", + .tp_basicsize = sizeof(struct pyrf_target), + .tp_dealloc = (destructor)pyrf_target__delete, + .tp_flags = Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, + .tp_doc = pyrf_target__doc, + .tp_init = (initproc)pyrf_target__init, +}; + +static int pyrf_target__setup_types(void) +{ + pyrf_target__type.tp_new = PyType_GenericNew; + return PyType_Ready(&pyrf_target__type); +} + +struct pyrf_record_opts { + PyObject_HEAD + + struct record_opts opts; +}; + +static int pyrf_record_opts__init(struct pyrf_record_opts *popts, + PyObject *args, PyObject *kwargs) +{ + static char *kwlist[] = { "target", "inherit_stat", "no_buffering", "no_inherit", + "no_inherit_set", "no_samples", "raw_samples", + "sample_address", "sample_phys_addr", "sample_data_page_size", + "sample_code_page_size", "sample_weight", "sample_time", + "sample_time_set", "sample_cpu", "sample_identifier", + "sample_data_src", "period", "period_set", "running_time", + "full_auxtrace", "auxtrace_snapshot_mode", + "auxtrace_snapshot_on_exit", "auxtrace_sample_mode", + "record_namespaces", "record_cgroup", "record_switch_events", + "record_switch_events_set", "all_kernel", "all_user", + "kernel_callchains", "user_callchains", "tail_synthesize", + "overwrite", "ignore_missing_thread", "strict_freq", "sample_id", + "no_bpf_event", "kcore", "text_poke", "build_id", "freq", + "mmap_pages", "auxtrace_mmap_pages", "user_freq", "branch_stack", + "sample_intr_regs", "sample_user_regs", "default_interval", + "user_interval", "auxtrace_snapshot_size", "auxtrace_snapshot_opts", + "auxtrace_sample_opts", "sample_transaction", "use_clockid", + "clockid", "clockid_res_ns", "nr_cblocks", "affinity", "mmap_flush", + "comp_level", "nr_threads_synthesize", "ctl_fd", "ctl_fd_ack", + "ctl_fd_close", "synth", "threads_spec", "threads_user_spec", + "off_cpu_thresh_ns", NULL }; + + if (!PyArg_ParseTupleAndKeywords(args, kwargs, + "|OppppppppppppppppppppppppppppppppppppppppIIIIIIIIIIssppiIiiiIIiipiisI", + kwlist, &(popts->opts.target), &(popts->opts.inherit_stat), + &(popts->opts.no_buffering), &(popts->opts.no_inherit), + &(popts->opts.no_inherit_set), &(popts->opts.no_samples), + &(popts->opts.raw_samples), &(popts->opts.sample_address), + &(popts->opts.sample_phys_addr), + &(popts->opts.sample_data_page_size), + &(popts->opts.sample_code_page_size), + &(popts->opts.sample_weight), + &(popts->opts.sample_time), &(popts->opts.sample_time_set), + &(popts->opts.sample_cpu), + &(popts->opts.sample_identifier), + &(popts->opts.sample_data_src), &(popts->opts.period), + &(popts->opts.period_set), &(popts->opts.running_time), + &(popts->opts.full_auxtrace), + &(popts->opts.auxtrace_snapshot_mode), + &(popts->opts.auxtrace_snapshot_on_exit), + &(popts->opts.auxtrace_sample_mode), + &(popts->opts.record_namespaces), + &(popts->opts.record_cgroup), + &(popts->opts.record_switch_events), + &(popts->opts.record_switch_events_set), + &(popts->opts.all_kernel), &(popts->opts.all_user), + &(popts->opts.kernel_callchains), + &(popts->opts.user_callchains), + &(popts->opts.tail_synthesize), + &(popts->opts.overwrite), + &(popts->opts.ignore_missing_thread), + &(popts->opts.strict_freq), &(popts->opts.sample_id), + &(popts->opts.no_bpf_event), &(popts->opts.kcore), + &(popts->opts.text_poke), &(popts->opts.build_id), + &(popts->opts.freq), &(popts->opts.mmap_pages), + &(popts->opts.auxtrace_mmap_pages), + &(popts->opts.user_freq), + &(popts->opts.branch_stack), + &(popts->opts.sample_intr_regs), + &(popts->opts.sample_user_regs), + &(popts->opts.default_interval), + &(popts->opts.user_interval), + &(popts->opts.auxtrace_snapshot_size), + &(popts->opts.auxtrace_snapshot_opts), + &(popts->opts.auxtrace_sample_opts), + &(popts->opts.sample_transaction), + &(popts->opts.use_clockid), + &(popts->opts.clockid), &(popts->opts.clockid_res_ns), + &(popts->opts.nr_cblocks), &(popts->opts.affinity), + &(popts->opts.mmap_flush), &(popts->opts.comp_level), + &(popts->opts.nr_threads_synthesize), + &(popts->opts.ctl_fd), + &(popts->opts.ctl_fd_ack), &(popts->opts.ctl_fd_close), + &(popts->opts.synth), &(popts->opts.threads_spec), + &(popts->opts.threads_user_spec), + &(popts->opts.off_cpu_thresh_ns))) + return -1; + + return 0; +} + +static const char pyrf_record_opts__doc[] = PyDoc_STR("perf record_opts object."); + +static void pyrf_record_opts__delete(struct pyrf_record_opts *perf_record_opts) +{ + Py_TYPE(perf_record_opts)->tp_free((PyObject *)perf_record_opts); +} + +static PyTypeObject pyrf_record_opts__type = { + PyVarObject_HEAD_INIT(NULL, 0) + .tp_name = "perf.record_opts", + .tp_basicsize = sizeof(struct pyrf_record_opts), + .tp_dealloc = (destructor)pyrf_record_opts__delete, + .tp_flags = Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, + .tp_doc = pyrf_record_opts__doc, + .tp_init = (initproc)pyrf_record_opts__init, +}; + +static int pyrf_record_opts__setup_types(void) +{ + pyrf_record_opts__type.tp_new = PyType_GenericNew; + return PyType_Ready(&pyrf_record_opts__type); +} + struct pyrf_evlist { PyObject_HEAD @@ -1263,7 +1426,7 @@ static PyObject *pyrf_evlist__close(struct pyrf_evlist *pevlist) return Py_None; } -static PyObject *pyrf_evlist__config(struct pyrf_evlist *pevlist) +static PyObject *pyrf_evlist__config(struct pyrf_evlist *pevlist, PyObject *args, PyObject *kwargs) { struct record_opts opts = { .sample_time = true, @@ -1281,8 +1444,17 @@ static PyObject *pyrf_evlist__config(struct pyrf_evlist *pevlist) .no_buffering = true, .no_inherit = true, }; + PyObject *popts = NULL; + static char *kwlist[] = { "record_opts", NULL }; struct evlist *evlist = &pevlist->evlist; + if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|O", kwlist, + &popts)) + return NULL; + + if (popts) + opts = ((struct pyrf_record_opts *) popts)->opts; + evlist__config(evlist, &opts, &callchain_param); Py_INCREF(Py_None); return Py_None; @@ -1354,7 +1526,7 @@ static PyMethodDef pyrf_evlist__methods[] = { { .ml_name = "config", .ml_meth = (PyCFunction)pyrf_evlist__config, - .ml_flags = METH_NOARGS, + .ml_flags = METH_VARARGS | METH_KEYWORDS, .ml_doc = PyDoc_STR("Apply default record options to the evlist.") }, { @@ -1718,7 +1890,9 @@ PyMODINIT_FUNC PyInit_perf(void) pyrf_evsel__setup_types() < 0 || pyrf_thread_map__setup_types() < 0 || pyrf_cpu_map__setup_types() < 0 || - pyrf_counts_values__setup_types() < 0) + pyrf_counts_values__setup_types() < 0 || + pyrf_target__setup_types() < 0 || + pyrf_record_opts__setup_types() < 0) return module; /* The page_size is placed in util object. */ @@ -1766,6 +1940,12 @@ PyMODINIT_FUNC PyInit_perf(void) Py_INCREF(&pyrf_counts_values__type); PyModule_AddObject(module, "counts_values", (PyObject *)&pyrf_counts_values__type); + Py_INCREF(&pyrf_target__type); + PyModule_AddObject(module, "target", (PyObject *)&pyrf_target__type); + + Py_INCREF(&pyrf_record_opts__type); + PyModule_AddObject(module, "record_opts", (PyObject *)&pyrf_record_opts__type); + dict = PyModule_GetDict(module); if (dict == NULL) goto error; -- 2.49.0 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 2/2] perf python: Add sampling.py as example for sampling 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani 2025-07-28 5:59 ` [PATCH 1/2] perf python: Add support for record_opts struct Gautam Menghani @ 2025-07-28 5:59 ` Gautam Menghani 2025-08-12 5:22 ` [PATCH 0/2] perf python: Add an " Gautam Menghani ` (2 subsequent siblings) 4 siblings, 0 replies; 12+ messages in thread From: Gautam Menghani @ 2025-07-28 5:59 UTC (permalink / raw) To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang Cc: Gautam Menghani, maddy, linux-perf-users, linux-kernel Add sampling.py - a python version of sampling.c to demonstrate sampling events. Sample output: ``` $ sudo ./sampling.py libperf: mmap_per_cpu: nr cpu values 8 nr threads 1 libperf: idx 0: mmapping fd 3 libperf: idx 1: mmapping fd 4 libperf: idx 2: mmapping fd 5 libperf: idx 3: mmapping fd 6 libperf: idx 4: mmapping fd 7 libperf: idx 5: mmapping fd 8 libperf: idx 6: mmapping fd 9 libperf: idx 7: mmapping fd 10 cpu: 0 pid: 4168 tid: 4168 ip: 0x7fb274150c15 period: 345012 cpu: 1 pid: 57137 tid: 57137 ip: 0xffffffffc0745a2c period: 286666 cpu: 2 pid: 6247 tid: 6247 ip: 0x7fe10a883988 period: 391180 cpu: 3 pid: 0 tid: 0 ip: 0xffffffffa5413e67 period: 245301 cpu: 4 pid: 4168 tid: 4194 ip: 0xffffffffc086a1cc period: 269605 cpu: 5 pid: 0 tid: 0 ip: 0xffffffffa53bfcb1 period: 90978 cpu: 6 pid: 0 tid: 0 ip: 0xffffffffa6469333 period: 309112 ``` Signed-off-by: Gautam Menghani <gautam@linux.ibm.com> --- tools/perf/python/sampling.py | 49 +++++++++++++++++++++++++++++++++++ 1 file changed, 49 insertions(+) create mode 100755 tools/perf/python/sampling.py diff --git a/tools/perf/python/sampling.py b/tools/perf/python/sampling.py new file mode 100755 index 000000000000..b5f4e1362c63 --- /dev/null +++ b/tools/perf/python/sampling.py @@ -0,0 +1,49 @@ +#!/usr/bin/env python3 +# SPDX-License-Identifier: GPL-2.0 +# -*- python -*- +# -*- coding: utf-8 -*- + +import perf +import time + +def main(): + cpus = perf.cpu_map() + threads = perf.thread_map(-1) + + evlist = perf.parse_events('cpu-cycles', cpus, threads) + + tgt = perf.target(uses_mmap = True, default_per_cpu = True) + opts = perf.record_opts(freq=1000, target=tgt, sample_time=True, + sample_cpu=True, no_buffering=True, no_inherit=True) + for ev in evlist: + ev.tracking = False + ev.read_format = 0 + ev.sample_type = perf.SAMPLE_IP|perf.SAMPLE_TID|perf.SAMPLE_CPU|perf.SAMPLE_PERIOD + evlist.config(opts) + + evlist.open() + evlist.mmap() + + evlist.enable() + time.sleep(2) + evlist.disable() + + done = False + while done is False: + for cpu in cpus: + event = evlist.read_on_cpu(cpu) + if event is None: + done = True + break + + if not isinstance(event, perf.sample_event): + continue + + print(f"cpu: {event.sample_cpu:<3} pid: {event.sample_pid:<6} " + f"tid: {event.sample_tid:<6} ip: {hex(event.sample_ip):<20} " + f"period: {event.sample_period:<20}") + + evlist.close() + +if __name__ == '__main__': + main() -- 2.49.0 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani 2025-07-28 5:59 ` [PATCH 1/2] perf python: Add support for record_opts struct Gautam Menghani 2025-07-28 5:59 ` [PATCH 2/2] perf python: Add sampling.py as example for sampling Gautam Menghani @ 2025-08-12 5:22 ` Gautam Menghani 2025-08-26 7:07 ` Gautam Menghani 2025-09-16 13:30 ` Gautam Menghani 4 siblings, 0 replies; 12+ messages in thread From: Gautam Menghani @ 2025-08-12 5:22 UTC (permalink / raw) To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang Cc: maddy, linux-perf-users, linux-kernel Hello, Please do review this series and let me know if any changes are needed. Thanks, Gautam ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani ` (2 preceding siblings ...) 2025-08-12 5:22 ` [PATCH 0/2] perf python: Add an " Gautam Menghani @ 2025-08-26 7:07 ` Gautam Menghani 2025-09-16 13:30 ` Gautam Menghani 4 siblings, 0 replies; 12+ messages in thread From: Gautam Menghani @ 2025-08-26 7:07 UTC (permalink / raw) To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang Cc: maddy, linux-perf-users, linux-kernel Hi, Gentle ping. Please review this series and let me know if any changes are needed. Thanks, Gautam ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani ` (3 preceding siblings ...) 2025-08-26 7:07 ` Gautam Menghani @ 2025-09-16 13:30 ` Gautam Menghani 2025-09-16 19:25 ` Arnaldo Carvalho de Melo 4 siblings, 1 reply; 12+ messages in thread From: Gautam Menghani @ 2025-09-16 13:30 UTC (permalink / raw) To: peterz, mingo, acme, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang Cc: maddy, linux-perf-users, linux-kernel Hi Ian/Arnaldo, Can you please review this series and let me know if any changes are needed? Thanks, Gautam ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-16 13:30 ` Gautam Menghani @ 2025-09-16 19:25 ` Arnaldo Carvalho de Melo 2025-09-16 20:07 ` Ian Rogers 0 siblings, 1 reply; 12+ messages in thread From: Arnaldo Carvalho de Melo @ 2025-09-16 19:25 UTC (permalink / raw) To: Gautam Menghani Cc: peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, irogers, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Tue, Sep 16, 2025 at 07:00:48PM +0530, Gautam Menghani wrote: > Hi Ian/Arnaldo, > > Can you please review this series and let me know if any changes are > needed? Looking at it now, sry for the delay, - Arnaldo ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-16 19:25 ` Arnaldo Carvalho de Melo @ 2025-09-16 20:07 ` Ian Rogers 2025-09-17 12:37 ` Arnaldo Carvalho de Melo 0 siblings, 1 reply; 12+ messages in thread From: Ian Rogers @ 2025-09-16 20:07 UTC (permalink / raw) To: Arnaldo Carvalho de Melo Cc: Gautam Menghani, peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Tue, Sep 16, 2025 at 12:25 PM Arnaldo Carvalho de Melo <acme@kernel.org> wrote: > > On Tue, Sep 16, 2025 at 07:00:48PM +0530, Gautam Menghani wrote: > > Hi Ian/Arnaldo, > > > > Can you please review this series and let me know if any changes are > > needed? > > Looking at it now, sry for the delay, I think the patches look good. I'm a little concerned that the python APIs are a chance to do something better than the C APIs that have evolved. For example, we removed UID out of target recently [1] as the BPF alternative was better. Had this patch come earlier then it seems likely we'd have had target with UIDs. I wonder rather than having a kwlist of: + static char *kwlist[] = { "target", "inherit_stat", "no_buffering", "no_inherit", + "no_inherit_set", "no_samples", "raw_samples", + "sample_address", "sample_phys_addr", "sample_data_page_size", + "sample_code_page_size", "sample_weight", "sample_time", + "sample_time_set", "sample_cpu", "sample_identifier", + "sample_data_src", "period", "period_set", "running_time", + "full_auxtrace", "auxtrace_snapshot_mode", + "auxtrace_snapshot_on_exit", "auxtrace_sample_mode", + "record_namespaces", "record_cgroup", "record_switch_events", + "record_switch_events_set", "all_kernel", "all_user", + "kernel_callchains", "user_callchains", "tail_synthesize", + "overwrite", "ignore_missing_thread", "strict_freq", "sample_id", + "no_bpf_event", "kcore", "text_poke", "build_id", "freq", + "mmap_pages", "auxtrace_mmap_pages", "user_freq", "branch_stack", + "sample_intr_regs", "sample_user_regs", "default_interval", + "user_interval", "auxtrace_snapshot_size", "auxtrace_snapshot_opts", + "auxtrace_sample_opts", "sample_transaction", "use_clockid", + "clockid", "clockid_res_ns", "nr_cblocks", "affinity", "mmap_flush", + "comp_level", "nr_threads_synthesize", "ctl_fd", "ctl_fd_ack", + "ctl_fd_close", "synth", "threads_spec", "threads_user_spec", + "off_cpu_thresh_ns", NULL }; but then just using this subset: + opts = perf.record_opts(freq=1000, target=tgt, sample_time=True, + sample_cpu=True, no_buffering=True, no_inherit=True) The kwlist should be kept to just those necessary values for the example to work? I kind of see this as Arnaldo's baby, so he may just want everything, so this needn't be a blocker. Bigger picture I wonder about migrating the `perf script` code to just being regular python programs like the example here. I sent out deprecating the libperl code to this ends (looking for reviews): https://lore.kernel.org/linux-perf-users/20250908181918.3533480-1-irogers@google.com/ The issue is that `perf script` being the main thread inhibits things like textual running until trace_end. This means we can't do things like incremental loading support. We may want to make the perf events support something like an asyncio interface for that. Refactoring that support will likely raise backward compatibility concerns. It'd be a really nice thing to do as the API has some fairly major overheads like turning everything in a sample into a Dict whether needed or not: https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/util/scripting-engines/trace-event-python.c#n838 I mention this just to say why I'd like to minimize the API when possible. Thanks, Ian [1] https://lore.kernel.org/r/20250604174545.2853620-10-irogers@google.com ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-16 20:07 ` Ian Rogers @ 2025-09-17 12:37 ` Arnaldo Carvalho de Melo 2025-09-17 15:29 ` Ian Rogers 0 siblings, 1 reply; 12+ messages in thread From: Arnaldo Carvalho de Melo @ 2025-09-17 12:37 UTC (permalink / raw) To: Ian Rogers Cc: Gautam Menghani, peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Tue, Sep 16, 2025 at 01:07:43PM -0700, Ian Rogers wrote: > On Tue, Sep 16, 2025 at 12:25 PM Arnaldo Carvalho de Melo > <acme@kernel.org> wrote: > > > > On Tue, Sep 16, 2025 at 07:00:48PM +0530, Gautam Menghani wrote: > > > Hi Ian/Arnaldo, > > > > > > Can you please review this series and let me know if any changes are > > > needed? > > > > Looking at it now, sry for the delay, > > I think the patches look good. I'm a little concerned that the python > APIs are a chance to do something better than the C APIs that have > evolved. For example, we removed UID out of target recently [1] as the > BPF alternative was better. Had this patch come earlier then it seems > likely we'd have had target with UIDs. I wonder rather than having a > kwlist of: > > + static char *kwlist[] = { "target", "inherit_stat", "no_buffering", > "no_inherit", > + "no_inherit_set", "no_samples", "raw_samples", > + "sample_address", "sample_phys_addr", "sample_data_page_size", > + "sample_code_page_size", "sample_weight", "sample_time", > + "sample_time_set", "sample_cpu", "sample_identifier", > + "sample_data_src", "period", "period_set", "running_time", > + "full_auxtrace", "auxtrace_snapshot_mode", > + "auxtrace_snapshot_on_exit", "auxtrace_sample_mode", > + "record_namespaces", "record_cgroup", "record_switch_events", > + "record_switch_events_set", "all_kernel", "all_user", > + "kernel_callchains", "user_callchains", "tail_synthesize", > + "overwrite", "ignore_missing_thread", "strict_freq", "sample_id", > + "no_bpf_event", "kcore", "text_poke", "build_id", "freq", > + "mmap_pages", "auxtrace_mmap_pages", "user_freq", "branch_stack", > + "sample_intr_regs", "sample_user_regs", "default_interval", > + "user_interval", "auxtrace_snapshot_size", "auxtrace_snapshot_opts", > + "auxtrace_sample_opts", "sample_transaction", "use_clockid", > + "clockid", "clockid_res_ns", "nr_cblocks", "affinity", "mmap_flush", > + "comp_level", "nr_threads_synthesize", "ctl_fd", "ctl_fd_ack", > + "ctl_fd_close", "synth", "threads_spec", "threads_user_spec", > + "off_cpu_thresh_ns", NULL }; > > but then just using this subset: > > + opts = perf.record_opts(freq=1000, target=tgt, sample_time=True, > + sample_cpu=True, no_buffering=True, > no_inherit=True) > > The kwlist should be kept to just those necessary values for the > example to work? I kind of see this as Arnaldo's baby, so he may just > want everything, so this needn't be a blocker. > > Bigger picture I wonder about migrating the `perf script` code to just > being regular python programs like the example here. You mean: acme@number:~/git/perf-tools-next$ ls -la tools/perf/scripts/python/ total 452 drwxr-xr-x. 1 acme acme 902 Aug 20 14:18 . drwxr-xr-x. 1 acme acme 30 Sep 17 09:27 .. -rwxr-xr-x. 1 acme acme 11865 Aug 20 14:06 arm-cs-trace-disasm.py drwxr-xr-x. 1 acme acme 1640 Aug 20 14:18 bin -rw-r--r--. 1 acme acme 2461 Apr 16 10:06 check-perf-trace.py -rw-r--r--. 1 acme acme 7923 Apr 16 10:06 compaction-times.py -rw-r--r--. 1 acme acme 7497 Apr 16 10:06 event_analyzing_sample.py -rwxr-xr-x. 1 acme acme 157369 Aug 20 14:18 exported-sql-viewer.py -rw-r--r--. 1 acme acme 39845 Apr 16 10:06 export-to-postgresql.py -rw-r--r--. 1 acme acme 24671 Apr 16 10:06 export-to-sqlite.py -rw-r--r--. 1 acme acme 2173 Apr 16 10:06 failed-syscalls-by-pid.py -rwxr-xr-x. 1 acme acme 10377 Aug 20 14:18 flamegraph.py -rw-r--r--. 1 acme acme 1717 Apr 16 10:06 futex-contention.py -rw-r--r--. 1 acme acme 13302 Apr 16 10:06 gecko.py -rw-r--r--. 1 acme acme 14636 Apr 16 10:06 intel-pt-events.py -rw-r--r--. 1 acme acme 3395 Apr 16 10:06 libxed.py -rw-r--r--. 1 acme acme 4230 Aug 20 14:13 mem-phys-addr.py -rw-r--r--. 1 acme acme 15420 Aug 20 14:05 netdev-times.py -rwxr-xr-x. 1 acme acme 1833 Apr 16 10:06 net_dropmonitor.py -rwxr-xr-x. 1 acme acme 30683 Aug 20 14:05 parallel-perf.py drwxr-xr-x. 1 acme acme 34 Sep 17 09:27 Perf-Trace-Util -rw-r--r--. 1 acme acme 4509 Apr 16 10:06 powerpc-hcalls.py -rw-r--r--. 1 acme acme 12095 Apr 16 10:06 sched-migration.py -rw-r--r--. 1 acme acme 2183 Apr 16 10:06 sctop.py -rwxr-xr-x. 1 acme acme 4408 Apr 16 10:06 stackcollapse.py -rw-r--r--. 1 acme acme 2444 Apr 16 10:06 stat-cpi.py -rw-r--r--. 1 acme acme 2055 Apr 16 10:06 syscall-counts-by-pid.py -rw-r--r--. 1 acme acme 1673 Apr 16 10:06 syscall-counts.py -rwxr-xr-x. 1 acme acme 34014 Apr 16 10:06 task-analyzer.py acme@number:~/git/perf-tools-next$ And then make: acme@number:~/git/perf-tools-next$ perf script -l List of available trace scripts: compaction-times [-h] [-u] [-p|-pv] [-t | [-m] [-fs] [-ms]] [pid|pid-range|comm-regex] display time taken by mm compaction event_analyzing_sample analyze all perf samples export-to-postgresql [database name] [columns] [calls] export perf data to a postgresql database export-to-sqlite [database name] [columns] [calls] export perf data to a sqlite3 database failed-syscalls-by-pid [comm] system-wide failed syscalls, by pid flamegraph create flame graphs futex-contention futext contention measurement gecko create firefox gecko profile json format from perf.data intel-pt-events print Intel PT Events including Power Events and PTWRITE mem-phys-addr resolve physical address samples net_dropmonitor display a table of dropped frames netdev-times [tx] [rx] [dev=] [debug] display a process of packet and processing time powerpc-hcalls sched-migration sched migration overview sctop [comm] [interval] syscall top stackcollapse produce callgraphs in short form for scripting use syscall-counts-by-pid [comm] system-wide syscall counts, by pid syscall-counts [comm] system-wide syscall counts task-analyzer analyze timings of tasks failed-syscalls [comm] system-wide failed syscalls rw-by-file <comm> r/w activity for a program, by file rw-by-pid system-wide r/w activity rwtop [interval] system-wide r/w top wakeup-latency system-wide min/max/avg wakeup latency acme@number:~/git/perf-tools-next$ And make: perf script rwtop Just call 'python PATH_TO_PYTHON_SCRIPTS/rwtop.py' transparently? That looks interesting indeed, that way we would stop linking with libpython, etc. I wonder if there are out of tree scripts using the current tools/perf/util/scripting-engines/trace-event-python.c mechanism... But even that can fallback to a python based mechanism, right? Import the script, if it has a given structure, use the new way, if not, call a glue that reads the events and feed to the old style code. Seems doable and would save code on the main perf binary and headaches with the libpython and libperl build processes. - Arnaldo > I sent out > deprecating the libperl code to this ends (looking for reviews): > https://lore.kernel.org/linux-perf-users/20250908181918.3533480-1-irogers@google.com/ > The issue is that `perf script` being the main thread inhibits things > like textual running until trace_end. This means we can't do things > like incremental loading support. We may want to make the perf events > support something like an asyncio interface for that. > > Refactoring that support will likely raise backward compatibility > concerns. It'd be a really nice thing to do as the API has some fairly > major overheads like turning everything in a sample into a Dict > whether needed or not: > https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/util/scripting-engines/trace-event-python.c#n838 > I mention this just to say why I'd like to minimize the API when possible. > > Thanks, > Ian > > [1] https://lore.kernel.org/r/20250604174545.2853620-10-irogers@google.com ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-17 12:37 ` Arnaldo Carvalho de Melo @ 2025-09-17 15:29 ` Ian Rogers 2025-09-17 16:41 ` Arnaldo Carvalho de Melo 0 siblings, 1 reply; 12+ messages in thread From: Ian Rogers @ 2025-09-17 15:29 UTC (permalink / raw) To: Arnaldo Carvalho de Melo Cc: Gautam Menghani, peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Wed, Sep 17, 2025 at 5:37 AM Arnaldo Carvalho de Melo <acme@kernel.org> wrote: > > On Tue, Sep 16, 2025 at 01:07:43PM -0700, Ian Rogers wrote: > > On Tue, Sep 16, 2025 at 12:25 PM Arnaldo Carvalho de Melo > > <acme@kernel.org> wrote: > > > > > > On Tue, Sep 16, 2025 at 07:00:48PM +0530, Gautam Menghani wrote: > > > > Hi Ian/Arnaldo, > > > > > > > > Can you please review this series and let me know if any changes are > > > > needed? > > > > > > Looking at it now, sry for the delay, > > > > I think the patches look good. I'm a little concerned that the python > > APIs are a chance to do something better than the C APIs that have > > evolved. For example, we removed UID out of target recently [1] as the > > BPF alternative was better. Had this patch come earlier then it seems > > likely we'd have had target with UIDs. I wonder rather than having a > > kwlist of: > > > > + static char *kwlist[] = { "target", "inherit_stat", "no_buffering", > > "no_inherit", > > + "no_inherit_set", "no_samples", "raw_samples", > > + "sample_address", "sample_phys_addr", "sample_data_page_size", > > + "sample_code_page_size", "sample_weight", "sample_time", > > + "sample_time_set", "sample_cpu", "sample_identifier", > > + "sample_data_src", "period", "period_set", "running_time", > > + "full_auxtrace", "auxtrace_snapshot_mode", > > + "auxtrace_snapshot_on_exit", "auxtrace_sample_mode", > > + "record_namespaces", "record_cgroup", "record_switch_events", > > + "record_switch_events_set", "all_kernel", "all_user", > > + "kernel_callchains", "user_callchains", "tail_synthesize", > > + "overwrite", "ignore_missing_thread", "strict_freq", "sample_id", > > + "no_bpf_event", "kcore", "text_poke", "build_id", "freq", > > + "mmap_pages", "auxtrace_mmap_pages", "user_freq", "branch_stack", > > + "sample_intr_regs", "sample_user_regs", "default_interval", > > + "user_interval", "auxtrace_snapshot_size", "auxtrace_snapshot_opts", > > + "auxtrace_sample_opts", "sample_transaction", "use_clockid", > > + "clockid", "clockid_res_ns", "nr_cblocks", "affinity", "mmap_flush", > > + "comp_level", "nr_threads_synthesize", "ctl_fd", "ctl_fd_ack", > > + "ctl_fd_close", "synth", "threads_spec", "threads_user_spec", > > + "off_cpu_thresh_ns", NULL }; > > > > but then just using this subset: > > > > + opts = perf.record_opts(freq=1000, target=tgt, sample_time=True, > > + sample_cpu=True, no_buffering=True, > > no_inherit=True) > > > > The kwlist should be kept to just those necessary values for the > > example to work? I kind of see this as Arnaldo's baby, so he may just > > want everything, so this needn't be a blocker. > > > > Bigger picture I wonder about migrating the `perf script` code to just > > being regular python programs like the example here. > > You mean: > > acme@number:~/git/perf-tools-next$ ls -la tools/perf/scripts/python/ > total 452 > drwxr-xr-x. 1 acme acme 902 Aug 20 14:18 . > drwxr-xr-x. 1 acme acme 30 Sep 17 09:27 .. > -rwxr-xr-x. 1 acme acme 11865 Aug 20 14:06 arm-cs-trace-disasm.py > drwxr-xr-x. 1 acme acme 1640 Aug 20 14:18 bin > -rw-r--r--. 1 acme acme 2461 Apr 16 10:06 check-perf-trace.py > -rw-r--r--. 1 acme acme 7923 Apr 16 10:06 compaction-times.py > -rw-r--r--. 1 acme acme 7497 Apr 16 10:06 event_analyzing_sample.py > -rwxr-xr-x. 1 acme acme 157369 Aug 20 14:18 exported-sql-viewer.py > -rw-r--r--. 1 acme acme 39845 Apr 16 10:06 export-to-postgresql.py > -rw-r--r--. 1 acme acme 24671 Apr 16 10:06 export-to-sqlite.py > -rw-r--r--. 1 acme acme 2173 Apr 16 10:06 failed-syscalls-by-pid.py > -rwxr-xr-x. 1 acme acme 10377 Aug 20 14:18 flamegraph.py > -rw-r--r--. 1 acme acme 1717 Apr 16 10:06 futex-contention.py > -rw-r--r--. 1 acme acme 13302 Apr 16 10:06 gecko.py > -rw-r--r--. 1 acme acme 14636 Apr 16 10:06 intel-pt-events.py > -rw-r--r--. 1 acme acme 3395 Apr 16 10:06 libxed.py > -rw-r--r--. 1 acme acme 4230 Aug 20 14:13 mem-phys-addr.py > -rw-r--r--. 1 acme acme 15420 Aug 20 14:05 netdev-times.py > -rwxr-xr-x. 1 acme acme 1833 Apr 16 10:06 net_dropmonitor.py > -rwxr-xr-x. 1 acme acme 30683 Aug 20 14:05 parallel-perf.py > drwxr-xr-x. 1 acme acme 34 Sep 17 09:27 Perf-Trace-Util > -rw-r--r--. 1 acme acme 4509 Apr 16 10:06 powerpc-hcalls.py > -rw-r--r--. 1 acme acme 12095 Apr 16 10:06 sched-migration.py > -rw-r--r--. 1 acme acme 2183 Apr 16 10:06 sctop.py > -rwxr-xr-x. 1 acme acme 4408 Apr 16 10:06 stackcollapse.py > -rw-r--r--. 1 acme acme 2444 Apr 16 10:06 stat-cpi.py > -rw-r--r--. 1 acme acme 2055 Apr 16 10:06 syscall-counts-by-pid.py > -rw-r--r--. 1 acme acme 1673 Apr 16 10:06 syscall-counts.py > -rwxr-xr-x. 1 acme acme 34014 Apr 16 10:06 task-analyzer.py > acme@number:~/git/perf-tools-next$ > > And then make: > > acme@number:~/git/perf-tools-next$ perf script -l > List of available trace scripts: > compaction-times [-h] [-u] [-p|-pv] [-t | [-m] [-fs] [-ms]] [pid|pid-range|comm-regex] display time taken by mm compaction > event_analyzing_sample analyze all perf samples > export-to-postgresql [database name] [columns] [calls] export perf data to a postgresql database > export-to-sqlite [database name] [columns] [calls] export perf data to a sqlite3 database > failed-syscalls-by-pid [comm] system-wide failed syscalls, by pid > flamegraph create flame graphs > futex-contention futext contention measurement > gecko create firefox gecko profile json format from perf.data > intel-pt-events print Intel PT Events including Power Events and PTWRITE > mem-phys-addr resolve physical address samples > net_dropmonitor display a table of dropped frames > netdev-times [tx] [rx] [dev=] [debug] display a process of packet and processing time > powerpc-hcalls > sched-migration sched migration overview > sctop [comm] [interval] syscall top > stackcollapse produce callgraphs in short form for scripting use > syscall-counts-by-pid [comm] system-wide syscall counts, by pid > syscall-counts [comm] system-wide syscall counts > task-analyzer analyze timings of tasks > failed-syscalls [comm] system-wide failed syscalls > rw-by-file <comm> r/w activity for a program, by file > rw-by-pid system-wide r/w activity > rwtop [interval] system-wide r/w top > wakeup-latency system-wide min/max/avg wakeup latency > acme@number:~/git/perf-tools-next$ > > And make: > > perf script rwtop > > Just call 'python PATH_TO_PYTHON_SCRIPTS/rwtop.py' transparently? Yeah, that's it. The perf script libpython stuff is just providing trace_begin, process_event and trace_end. Using the sampling mechanism, as shown in Gautum's patches, it is pretty easy to migrate them to being stand alone bits of python. > That looks interesting indeed, that way we would stop linking with > libpython, etc. > > I wonder if there are out of tree scripts using the current tools/perf/util/scripting-engines/trace-event-python.c > mechanism... > > But even that can fallback to a python based mechanism, right? I think so. Like I said about the use of a Dict for process_event, we may want to streamline things so there is a tension with what the API should be and compatibility. We can always have 2 APIs and try to deprecate one of them. > Import the script, if it has a given structure, use the new way, if not, > call a glue that reads the events and feed to the old style code. > > Seems doable and would save code on the main perf binary and headaches > with the libpython and libperl build processes. So I see this for libpython, and I think it'd be pretty cool if we could have things work like this for say "perf script ilist" and Alice's textual flamegraph work. I worry what the work for libperl would be like and whether it is worth it (hence sending the patch to at least start to make it opt-in rather than opt-out). Do you need my tags for these changes or wdyt about making the kwlist/API surface smaller? Thanks! Ian > - Arnaldo > > > I sent out > > deprecating the libperl code to this ends (looking for reviews): > > https://lore.kernel.org/linux-perf-users/20250908181918.3533480-1-irogers@google.com/ > > The issue is that `perf script` being the main thread inhibits things > > like textual running until trace_end. This means we can't do things > > like incremental loading support. We may want to make the perf events > > support something like an asyncio interface for that. > > > > Refactoring that support will likely raise backward compatibility > > concerns. It'd be a really nice thing to do as the API has some fairly > > major overheads like turning everything in a sample into a Dict > > whether needed or not: > > https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/util/scripting-engines/trace-event-python.c#n838 > > I mention this just to say why I'd like to minimize the API when possible. > > > > Thanks, > > Ian > > > > [1] https://lore.kernel.org/r/20250604174545.2853620-10-irogers@google.com ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-17 15:29 ` Ian Rogers @ 2025-09-17 16:41 ` Arnaldo Carvalho de Melo 2025-09-17 17:26 ` Ian Rogers 0 siblings, 1 reply; 12+ messages in thread From: Arnaldo Carvalho de Melo @ 2025-09-17 16:41 UTC (permalink / raw) To: Ian Rogers Cc: Gautam Menghani, peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Wed, Sep 17, 2025 at 08:29:24AM -0700, Ian Rogers wrote: > On Wed, Sep 17, 2025 at 5:37 AM Arnaldo Carvalho de Melo <acme@kernel.org> wrote: > > And make: > > perf script rwtop > > Just call 'python PATH_TO_PYTHON_SCRIPTS/rwtop.py' transparently? > Yeah, that's it. The perf script libpython stuff is just providing > trace_begin, process_event and trace_end. Using the sampling > mechanism, as shown in Gautum's patches, it is pretty easy to migrate > them to being stand alone bits of python. > > That looks interesting indeed, that way we would stop linking with > > libpython, etc. > > I wonder if there are out of tree scripts using the current tools/perf/util/scripting-engines/trace-event-python.c > > mechanism... > > But even that can fallback to a python based mechanism, right? > I think so. Like I said about the use of a Dict for process_event, we > may want to streamline things so there is a tension with what the API > should be and compatibility. We can always have 2 APIs and try to > deprecate one of them. > > Import the script, if it has a given structure, use the new way, if not, > > call a glue that reads the events and feed to the old style code. > > Seems doable and would save code on the main perf binary and headaches > > with the libpython and libperl build processes. > So I see this for libpython, and I think it'd be pretty cool if we > could have things work like this for say "perf script ilist" and I think we could even try to, not having a builtin-ilist.c convert 'perf ilist' to 'perf script ilist' automagically. > Alice's textual flamegraph work. I worry what the work for libperl > would be like and whether it is worth it (hence sending the patch to > at least start to make it opt-in rather than opt-out). Maybe we should be a tad more cautious and start with emitting a warning that "libperl is deprecated (holler if you disagree if you use it!)" and then make it opt-in, and then remove it. > Do you need my tags for these changes or wdyt about making the > kwlist/API surface smaller? Lemme look at the original post... - Arnaldo ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/2] perf python: Add an example for sampling 2025-09-17 16:41 ` Arnaldo Carvalho de Melo @ 2025-09-17 17:26 ` Ian Rogers 0 siblings, 0 replies; 12+ messages in thread From: Ian Rogers @ 2025-09-17 17:26 UTC (permalink / raw) To: Arnaldo Carvalho de Melo Cc: Gautam Menghani, peterz, mingo, namhyung, mark.rutland, alexander.shishkin, jolsa, adrian.hunter, kan.liang, maddy, linux-perf-users, linux-kernel On Wed, Sep 17, 2025 at 9:41 AM Arnaldo Carvalho de Melo <acme@kernel.org> wrote: > > On Wed, Sep 17, 2025 at 08:29:24AM -0700, Ian Rogers wrote: > > On Wed, Sep 17, 2025 at 5:37 AM Arnaldo Carvalho de Melo <acme@kernel.org> wrote: > > > And make: > > > > perf script rwtop > > > > Just call 'python PATH_TO_PYTHON_SCRIPTS/rwtop.py' transparently? > > > Yeah, that's it. The perf script libpython stuff is just providing > > trace_begin, process_event and trace_end. Using the sampling > > mechanism, as shown in Gautum's patches, it is pretty easy to migrate > > them to being stand alone bits of python. > > > > That looks interesting indeed, that way we would stop linking with > > > libpython, etc. > > > > I wonder if there are out of tree scripts using the current tools/perf/util/scripting-engines/trace-event-python.c > > > mechanism... > > > > But even that can fallback to a python based mechanism, right? > > > I think so. Like I said about the use of a Dict for process_event, we > > may want to streamline things so there is a tension with what the API > > should be and compatibility. We can always have 2 APIs and try to > > deprecate one of them. > > > > Import the script, if it has a given structure, use the new way, if not, > > > call a glue that reads the events and feed to the old style code. > > > > Seems doable and would save code on the main perf binary and headaches > > > with the libpython and libperl build processes. > > > So I see this for libpython, and I think it'd be pretty cool if we > > could have things work like this for say "perf script ilist" and > > I think we could even try to, not having a builtin-ilist.c convert > 'perf ilist' to 'perf script ilist' automagically. > > > Alice's textual flamegraph work. I worry what the work for libperl > > would be like and whether it is worth it (hence sending the patch to > > at least start to make it opt-in rather than opt-out). > > Maybe we should be a tad more cautious and start with emitting a warning > that "libperl is deprecated (holler if you disagree if you use it!)" and > then make it opt-in, and then remove it. Not sure I follow, let's move the conversation to that patch. > > Do you need my tags for these changes or wdyt about making the > > kwlist/API surface smaller? > > Lemme look at the original post... Thanks, Ian > - Arnaldo ^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2025-09-17 17:27 UTC | newest] Thread overview: 12+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2025-07-28 5:59 [PATCH 0/2] perf python: Add an example for sampling Gautam Menghani 2025-07-28 5:59 ` [PATCH 1/2] perf python: Add support for record_opts struct Gautam Menghani 2025-07-28 5:59 ` [PATCH 2/2] perf python: Add sampling.py as example for sampling Gautam Menghani 2025-08-12 5:22 ` [PATCH 0/2] perf python: Add an " Gautam Menghani 2025-08-26 7:07 ` Gautam Menghani 2025-09-16 13:30 ` Gautam Menghani 2025-09-16 19:25 ` Arnaldo Carvalho de Melo 2025-09-16 20:07 ` Ian Rogers 2025-09-17 12:37 ` Arnaldo Carvalho de Melo 2025-09-17 15:29 ` Ian Rogers 2025-09-17 16:41 ` Arnaldo Carvalho de Melo 2025-09-17 17:26 ` Ian Rogers
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).