pytest.toml without section header leads to silently ignored options

[The-Compiler]

With this pytest.toml:

addopts = "-v"

pytest silently ignores the option without any sort of warning/error.

This is different to a pytest.ini, where

addopts = -v

leads to ERROR: .../pytest.ini:1: no section header defined.

One would think that a [pytest] header is not neaded at all, because sections (aka tables) are completely optional in toml, and with the config being in a pytest.toml (and not pyproject.toml), the [pytest] header is optional.

IMHO, we should go that way, and read the toplevel option without a section (and possibly even do so for pytest.ini, at least this is supported by configparser, not sure about inicfg).

Alternatively, pytest should warn/error about there being toplevel options it did not read.

---

Additionally, a similar issue exists with both TOML and ini if there is a typo in the section name. Again in pyproject.toml this makes perfect sense, but there is no reason there should be a [pytestt] section in either pytest.toml or pytest.ini. If I remember correctly there have been plugins (don't remember which) that have their own section in there and then read that out manually (without pytest's infrastructure), so maybe this should be a warning rather than an error.

[vismaytiwari]

Hi! I was able to reproduce this locally and opened a PR with a focused fix: #14646

The issue happens because pytest.toml is detected as the config file, but top-level options are ignored when there is no [pytest] table. The PR reads top-level non-table options from pytest.toml / .pytest.toml when [pytest] is absent.

I added a regression test, docs, and a changelog fragment. I also verified it with a tiny manual repro where top-level addopts = ["-v"] is now honored, plus the relevant config test suite.

[nicoddemus]

I would prefer that we don't special case the format for pytest.toml, e.g., do not allow top level non-table options to be considered. We should always require the [pytest] table, my rationale being that the convenience is small compared to having to implement/maintain/document the difference.

However, I do agree that we should error out if there is no [pytest] table in a pytest.toml file: it is pytest specific, so a missing [pytest] table is certainly an error. Of course, we should not error out when using pyproject.toml as root but it is a pytest table, as pyproject.toml is not pytest exclusive.

[bluetech]

I agree with @nicoddemus about requiring [pytest] for consistency. Maybe if we went with it from the start, but the inertia now is against it.

However, I do agree that we should error out if there is no [pytest] table in a pytest.toml file

I'm not sure about this. The mere existence of pytest.toml has an effect, so it's reasonable to have an empty one IMO.

Alternatively, pytest should warn/error about there being toplevel options it did not read.

This sounds good to me to protect against typos. We can maybe gate this with --strict-ini, though that would somewhat defeat the purpose.

There is a question of whether there is a use case for anyone who is not core pytest to add sections to pytest.toml. I'm thinking mostly plugins. Currently plugins configure in [pytest] but it would be nice if they had their own sections. I remember @RonnyPfannschmidt has plans for this. How this affects this discussion depends on how it's implemented: if the plugin registers its section with pytest core, then pytest core can then allow it. If not, maybe it can use a sanctioned plugin prefix e.g. [plugin.django], like tool in pyproject.toml.

[nicoddemus]

I'm not sure about this. The mere existence of pytest.toml has an effect, so it's reasonable to have an empty one IMO.

Good point, perhaps if there are top-level options and it does not contain a [pytest] table?

[vismaytiwari]

Thanks, that makes sense.

Would you like me to update the PR to keep [pytest] required, but report top-level options in pytest.toml / .pytest.toml when there is no [pytest] table?

So an empty pytest.toml would still be allowed, but a file like:

toml
addopts = "-v"

would no longer be silently ignored.