Fix warning about non-ascii warnings even when they are ascii

[nicoddemus]

Fix #2809

[nicoddemus]

It was always issue the "unicode message non-ascii compatible" because warns_msg.args is a tuple, which always compares as different than a list.

[coveralls]

Coverage remained the same at 92.232% when pulling 94ae943 on nicoddemus:issue-2809 into 6690b8a on pytest-dev:master.

[RonnyPfannschmidt]

interesting find ^^

[nicoddemus]

We didn't realize at the time, but this is incorrect, safe_str will return utf-8 encoded bytes, which is not ascii.

[coveralls]

Coverage remained the same at 92.232% when pulling e49db9c on nicoddemus:issue-2809 into 6690b8a on pytest-dev:master.

[RonnyPfannschmidt]

i fear this one could count as a breaking change as output characters are removed now

[nicoddemus]

You are absolutely right, thanks. Rebased.

[coveralls]

Coverage remained the same at 92.5% when pulling fbb9e93 on nicoddemus:issue-2809 into d132c50 on pytest-dev:features.

[nicoddemus]

Can we merge this @RonnyPfannschmidt ?

[RonnyPfannschmidt]

@nicoddemus please add the changelog entry prefix for a deprecation/removal as this one is a bigger one

alltho i still don't see why we cant just escape

[nicoddemus]

alltho i still don't see why we cant just escape

What do you mean? I'm open to suggestions here.

[RonnyPfannschmidt]

@nicoddemus i dont see why we dont keep escaping as safe-string, why remove the unicode escapes

as far as i can tell it shouldnbt affect ascii, but it should affect non-ascii that way - just

i mean why not just use unicode-escape instead of safe_string for any unicode element?

[nicoddemus]

i dont see why we dont keep escaping as safe-string, why remove the unicode escapes

Fair enough, I should have mentioned this: if we use safe_str, then we get another warning:

UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
    unicode_warning = list(warn_msg.args) != new_args

safe_str returns utf-8 encoded bytes. This warning was not appearing before because the comparison was short-circuiting because of the tuple != list comparison, which didn't even try to do element-wise comparison because the types were different.

Because of this, new_args cannot be converted to unicode using ascii, which is what Python tries to do with the right side of the comparison.

[nicoddemus]

i mean why not just use unicode-escape instead of safe_string for any unicode element?

That's a good suggestion!

Used compat.ascii_escaped which does what you suggest.

Also renamed _ascii_escaped to ascii_escaped since it lives in the compat module which is internal anyway (all other functions in compat don't have a _ prefix as well).

[coveralls]

Coverage remained the same at 92.5% when pulling df6d5cd on nicoddemus:issue-2809 into d132c50 on pytest-dev:features.