Allow multiple use of `-m` option

[joaonc]

Looks like when we use -m more than once, only the last entry is used (on Linux/macOS, at least, haven't tested on Windows).

Ex:

pytest test_me.py -m x -m z
pytest test_me.py -m 'x or y' -m z

Both those examples will only run those tests with the mark z, b/c that's the last -m argument passed in. If we reverse the order, then the other marked tests will run.

Nowadays it's common for command line programs to take in multiple of the same arguments and they usually work in an additive way.
In this case, multiple -m usage would mean an or:

• -m x -m z = -m 'x or z'
• -m 'x or y' -m z = -m '(x or y) or z'

If this is not possible, then fail when using -m multiple times.

My use case is that I have a bash script that adds markers depending on arguments and environment variables and took me a while to figure this out.. having early failure would have saved a lot of time.

[RonnyPfannschmidt]

Both marks and keyword expressions currently use the last passed value

If added multiple times they should use and, not or

[symonk]

@RonnyPfannschmidt / @nicoddemus should we permit the -m on parsearg to 'append' (nargs='+') and then resolve the collection of arguments into one matchexpr, joining on 'and' ? not sure where -k would fit in here.

[RonnyPfannschmidt]

Both should use append, both need a reset arg, we should evaluate the expressions separately

[symonk]

apologies my comment was confusing :) I mean to keep them seperated, my thoughts are:

replace both -k / -m with a append action (nargs is not necessary)
resolve them into a joined string for each instance of -m and respectively -k (seperately ofcourse), joined with 'and'

for example:

pytest -m a or b -m c

matchexpr = 'a or b and c'

any more info into this 'reset arg'? not entirely what is meant there

thanks

[RonnyPfannschmidt]

@symonk instead of string joining, i would like to go the safe route of simply running a for loop over the expressions requiring an all,

reset options would clear the marker/keyword lists, so they could restart in case of addopts/automation putting in things one wants to ignore

[symonk]

thanks @RonnyPfannschmidt , I think selectuntil throws a bit of a spanner in the works here, still getting to grips with how this works but will try and implement as advised, can't think of a way other than for colitem in items for each keywordexpr to handle that gracefully

[symonk]

needs more of a discussion as likely a breaking change

[Sup3rGeo]

In the meantime would it work to try writing a plugin implementing the load_initial_conftest hook and implement your own custom additive marker option that you can pre-parse in the end create the final -m option?

[bluetech]

Some related discussion happened recently here: #7093.