I looked into the issue you told me about the other day, that one_line_regex.match_magic() doesn't work when called from within a function, but only when called from the top-level scope.
That is because accessing a variable that hasn't been defined results into a LOAD_GLOBAL instruction. But you assign to the caller's local scope. However, if the function is called from the top-level scope f_globals == f_locals. That is the reason it still worked in that scenario. So the fix is as simple as assigning to the callers global scope instead, and voila magic_match() works within function calls as well.
Of course this bears side effects like polluting the global scope, but I'm confident this is the only way to make it work (without a decorator rewriting the byte code). Also there is still a corner case left that seems to be impossible to tackle (without rewriting the byte code):
def foo():
match = None
magic_match(r'.*', 'foo')
print match
In that case FAST_LOAD is used to lookup the variable match and globals are ignored. However, assigning to locals wouldn't help either, because the value of the variable is looked up from the stack.
The other changes in this PR are just minor code cleanups (isolated in separate commits) to allow importing this file actually as module and passing additional parameters such as flags.
I looked into the issue you told me about the other day, that
one_line_regex.match_magic()doesn't work when called from within a function, but only when called from the top-level scope.That is because accessing a variable that hasn't been defined results into a
LOAD_GLOBALinstruction. But you assign to the caller's local scope. However, if the function is called from the top-level scopef_globals == f_locals. That is the reason it still worked in that scenario. So the fix is as simple as assigning to the callers global scope instead, and voilamagic_match()works within function calls as well.Of course this bears side effects like polluting the global scope, but I'm confident this is the only way to make it work (without a decorator rewriting the byte code). Also there is still a corner case left that seems to be impossible to tackle (without rewriting the byte code):
In that case
FAST_LOADis used to lookup the variablematchand globals are ignored. However, assigning to locals wouldn't help either, because the value of the variable is looked up from the stack.The other changes in this PR are just minor code cleanups (isolated in separate commits) to allow importing this file actually as module and passing additional parameters such as flags.