`_remote_debugging`: `write_sample` advances the timestamp before writing the sample

[maurycy]

Bug report

(found while thinking about how to address the code review from #157228)

The timestamp is advanced before the stack is built:

https://github.com/python/cpython/blob/b9546455ab0/Modules/_remote_debugging/binary_io_writer.c#L952-L953

If there's a failure, such as:

cpython/Lib/profiling/sampling/sample.py

Lines 180 to 184 in 763b6ed

	except (RuntimeError, UnicodeDecodeError, MemoryError, OSError):
	flush_pending()
	collector.collect_failed_sample()
	errors += 1
	prev_stack = None

Then the timestamps for the thread is shifted. That's a hard corruption for a delta-based format.

Reproduction

maurycy@gimel cpython-315 (main e580c88?) % cat repro.py
import _remote_debugging

w = _remote_debugging.BinaryWriter("/tmp/o.bin", 1000, 0)
try:
    w.write_sample([(0, [(1, 0, [(42, None, "f", None)])])], 5000)
except TypeError:
    pass
w.write_sample([(0, [(1, 0, [("a.py", None, "f", None)])])], 6000)
w.finalize()

class C:
    def collect(self, frames, timestamps_us=None, **kw):
        print(timestamps_us)

_remote_debugging.BinaryReader("/tmp/o.bin").replay(C())
maurycy@gimel cpython-315 (main e580c88?) % ./python.exe repro.py
[1000]

Another repro could involve _testcapi.set_nomemory, incorrect unicode etc.

CPython versions tested on:

CPython main branch, 3.15

Operating systems tested on:

macOS

Linked PRs

• gh-158550: _remote_debugging: Do not advance prev_timestamp when write_sample fails #158551