Missed lines with asyncio & context managers

[nedbat]

Originally reported by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

I have a rather odd problem I cannot reproduce in clean room.

I have a rather simple class and method:

https://gist.github.com/hynek/3ad620d07a7999f4daf5

Line 21 – or any line before the except – is not reported as covered. Even if it’s a print statement and see its output.

It seems to be related with the context manager because it's a) always the last line and b) when I rewrite it as https://gist.github.com/hynek/cfe2cc32f337e2676466 the coverage works just fine.

I’m sorry I can’t be more helpful but I’m confused as hell myself. The pool is from aiopg and seems to be semi-magical: https://github.com/aio-libs/aiopg/blob/master/aiopg/pool.py#L302 so dunno if that helps…

---

• Bitbucket: https://bitbucket.org/ned/coveragepy/issue/441

[nedbat]

@hynek any chance you've come up with more information about this? Can you provide explicit instructions for reproducing it, even if it's a particular hash in your repo with details of how to run the tests?

[nedbat]

Original comment by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

I was able to strip it down reasonably: https://ox.cx/stuff/sscce.tbz2

Preparations

You need a postgres DB that you have to configure in test.yml.

This DB needs a single table:

CREATE TABLE "public"."tokens" (
	"token" text NOT NULL COLLATE "default",
	"system" text NOT NULL COLLATE "default",
	"purpose" text NOT NULL COLLATE "default",
	"user_id" int4 NOT NULL,
	"created" timestamp(6) WITH TIME ZONE DEFAULT timezone('UTC'::text, now()),
	"valid" bool NOT NULL DEFAULT true
)
WITH (OIDS=FALSE);
ALTER TABLE "public"."tokens" ADD PRIMARY KEY ("token") NOT DEFERRABLE INITIALLY IMMEDIATE;

Reproduction

• create a Python 3.5 venv
• pip install -r dev-requirements.txt
• run make

You should be seeing src/tokens/aio.py 25 0 6 1 97% 34->35.

If you add a XXX in that line though, the test will explode.

---

Let me know if you need something.

[nedbat]

I haven't had a chance to reproduce this yet, but it sure looks like a duplicate of #146. Would you mind trying 4.1b1 to see if it is fixed? https://pypi.python.org/pypi/coverage/4.1b1

[nedbat]

Original comment by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

Sadly, 4.1b1 seems worse.

I’ve updated, did an coverage erase, and run coverage run -m pytest.

I’m getting the following output:

Name                     Stmts   Miss Branch BrPart  Cover   Missing
--------------------------------------------------------------------
src/tokens/__init__.py       3      0      0      0   100%
src/tokens/_common.py       14      0      2      0   100%
src/tokens/aio.py           26      0     14      3    92%   17->-5, 34->-23, 36->-23
src/tokens/rnd.py           20      0      4      0   100%
--------------------------------------------------------------------
TOTAL                       63      0     20      3    96%

• line 17 is the docstring of a class definition,
• line 34 is the context manager as before,
• and 36 is an except clause.

In htmlcov:

[nedbat]

Hmm, the class docstring is confusing me: I had this happen in code of mine, and the first change I made after b1 was to change it again. About the await's: it looks like I need to understand them better. I'll try your code.

[nedbat]

I'm understanding await and yield from better: looks like I shouldn't consider them exits from the function. Discussion here if you want to chime in: https://groups.google.com/d/topic/python-tulip/sdM8KKz3XWk/discussion

[nedbat]

Original comment by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

I’m on the ML and I saw the thread. Sadly my understanding of asyncio is still rather superficial ATM so I don’t have much to add. :(

[nedbat]

This will be fixed in 4.1b2.

[nedbat]

Original comment by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

I can confirm that the current default branch fixes all asyncio problems I was aware of! Thanks Ned!

[nedbat]

Original comment by Hynek Schlawack (Bitbucket: hynek, GitHub: hynek)

---

OK I’m afraid I’ve found another similar case:

try:
    foo
except Something:
   raise  # <-- can't  get this covered

If you wonder what that construct is good for: I want to whitelist certain exceptions that can pass unaltered but wrap all others.

Should I reopen or open a new issue?

[nedbat]

A new issue would be best, thanks.