In recent Python the following
>>> from ast import PyCF_ONLY_AST
>>> compile("'a'", 'whatever', 'exec', PyCF_ONLY_AST).body
In 3.6 it return
[<_ast.Expr at 0x10b7441d0>] # that contail Str('a')
While on master:
[]
This is inconveninent for alternative repl like IPython, where basically if the user is entering a single string, the result is nothing in Python 3.7+, while it _does_ return something on earlier Python [1].
The documentation of `compile` says:
> ... it can be 'exec' if source consists of a sequence of statements,
Which is not technically true any more as the first statement, if a string, will be removed.
What's happening here is that since Python 3.7 if the _first_ statement is actually an expression containing a lonely string it is assign to the module docstring. So that's basically assuming you are parsing a module, and that the docstring make sens in this context, while in a REPL you are parsing a sucesssion of statements, in which case there is no need for a docstring that make no sens in this context.
This is _usually_ not an issue, unless this lonely statement is also the last, and what the user wants to execute in a REPL, in which case it has no side effect.
I don't have any objection to the new behavior, though I was wondering if this kind of side effect was anticipated.
If that affect IPython, it will likely effect other alternative REPLs.
Thus, I believe it would be good to have this at least documented a tiny bit better and added in what's new, and potentially clarified in the `exec` docs.
I could argue that now the `exec` name may be a tiny bit unsuitable for the new behavior, and would love if this could be optional.
1: https://github.com/ipython/ipython/issues/11133#issuecomment-388591332
Fair enough that what's new include things about Module
> The first statement in their body is not considered as a docstring anymore.
Note that this sentence read backward to me. I understand what is meant because I know the new behavior. It might be good to clarify. potentially:
> The first statement in the `body` attribute of should not be considered the docstring of the module anymore, the `docstring` attribute is reserved for that.
Though the documentation of `compile()` does not say that `compile(...,'exec')` compile a module. It says:
> The mode argument specifies what kind of code must be compiled;
> it can be 'exec' if source consists of a sequence of statements
Which now is incorrect. I was expecting `compile(..., 'exec')` to return a Module with `None` or empty string as the docstring attribute – which is also a perfectly reasonable request.
I think that `compile` documentation should be changed to reflect what it does.
Or (but I see why this is un-reasonable) split add the `mode='module'` that has new behavior, while `exec` does not.