Does the position of @run_after matter during decoration?

[adamjstewart]

@alalazo

I was experimenting with adding an optional post-installation test to a package and noticed the following behavior.

@run_after('install')                                                       
@on_package_attributes(run_tests=True)                                      
def check_install(self):

works as expected. The test is run post-installation, but only when --run-tests is specified. However, if I do:

@on_package_attributes(run_tests=True)      
@run_after('install')                                                                                       
def check_install(self):

instead, the tests are always run regardless of whether or not I supply --run-tests.

[alalazo]

@adamjstewart That's because when you call @run_after or @run_before you register the function to be called back, for instance:

def run_after(*phases):
    """Registers a method of a package to be run after a given phase"""
    return PackageMeta.register_callback('run_after', *phases)

Now, if you "unroll" the second case to its semantically equivalent:

def check_install(self):
   pass

check = run_after('install')(check)  # Gets registered here
check = on_package_attribute(run_tests=True)(check)  # ... but also gets decorated after the registration

you see that the second decorator has no effect on the registered function. Maybe worth documenting?

[alalazo]

@adamjstewart I changed the title to reflect better what is being discussed here. Feel free to change it back if you disagree.

[adamjstewart]

I think it's a bit too complicated to document this one. There's no way to just make it work?

[adamjstewart]

@alalazo What if we changed the interface to:

@run_after('install', run_tests=True)                                                       
def check_install(self):

I'm not sure how the implementation details would work, but that would avoid a lot of unnecessary confusion.