remove ANSI escape sequences from the report messages

[keewis]

pytest has somewhat recently added support for coloring diffs of builtin structures like list or dict, which makes ANSI escape sequences appear in the messages pytest-reportlog writes to the log file.

Since that log file is intended to be machine-readable, I think reportlog should either strip the escape sequences from error messages, or pytest should add colors after the execution of the hook that reportlog registers to. I'd guess the former to be more realistic since I'd expect modifying pytest to that effect to be really tricky.

[RonnyPfannschmidt]

Im.a bit mixed about this as the coloring is helpful in a number of situations and ansi escape is a standard

[keewis]

Indeed, the visual clues are helpful (especially for humans), but usually make processing the messages programmatically much harder. Also, this is controlled by the --color option / the FORCE_COLOR env variable / the terminal capabilities, which makes it pretty hard to predict the output for programmatic use.

As an example, pytest may truncate the error message under certain circumstances, but in doing so does not take into account any invisible characters / the styling. This may result in it cutting somewhere in the middle of a escape sequence, or allowing the previous coloring / styling to bleed into any unstyled text following it (I didn't check yet if that has been reported already or not).

[RonnyPfannschmidt]

Sounds like we have multiple problems at gand once a closer look is taken

[nicoddemus]

I agree we strip the ANSI escape from the report log in general, with an option (defaulting to False) to keep them when desired -- I suspect most cases would like to see the report log without any escape sequences.

Perhaps this is something that should be improved on pytest itself: the Test and Collection reports should have two versions of their outputs: original (which may include ansi escapes or not) and escaped versions (guaranteed to never include ANSI escapes). Then consumers might opt-in to either depending on their use case.