lxml missing type annotation for XPath

[uglycoyote]

The following small script

from lxml import etree    
NameXPath = etree.XPath("Name/text()")

gives the error

test.py:3: error: "module" has no attribute "XPath"

which appears to be because the lxml typings are incomplete. I confirmed that this source does not mention XPath

https://github.com/python/typeshed/blob/master/third_party/3/lxml/etree.pyi

but I don't know enough about mypy or the types in lxml to propose a patch myself.

I asked a question about this on Stack Overflow (http://stackoverflow.com/questions/39382937/mypy-spurious-error-module-has-no-attribute-xpath-with-etree#39382937) which brought me here.

[Naruto0]

It seems that lmxl stubs are not complete. On code from lxml import html I have following error __main__.py:64: error: Module has no attribute 'html'

[gvanrossum]

Correct diagnosis. Can you or someone else submit a PR to improve the lxml stubs?

[Naruto0]

If @uglycoyote won't complain i call dibs on lxml typeshed.

[gvanrossum]

Go for it!

[uglycoyote]

Yes, naruto0, please do. I am only a casual user of lxml and don't really understand the types that it uses, hence my need for typeshed to help me. Thanks for offering. And thanks Guido for nudging this along.

[Naruto0]

Bump, i'm working on it, have been busy lately.

[underyx]

Here's a quick and friendly ping for @Naruto0

[gvanrossum]

Assuming @Naruto0 is still busy, would you (@underyx) want to submit a PR yourself?

[underyx]

Unfortunately my experience with mypy and type annotations can be summed up as 'tried it on a project, saw the lxml failures, decided to stop trialling mypy', so, just like the issue author, I don't think I'm qualified to take on the task.

Though, I am planning to look into the topic further in the coming weeks, so no idea if this'll stay the case.

[uglycoyote]

It would be great if there was a tool which helped someone explore the types that a library uses, to help someone make a correct pyi type information file without necessarily being an export on the source.

I understand that there is an existing tool to statically analyze sources and generate a stub pyi file (stubgen) but when I tried running "stubgen.py lxml" I got a file with exactly one line

def get_include(): ...

Also it seems unlikely to me that the static approach would work very well: since the sources don't declare the types of any function or variable, how on earth would a static analyzer know what type a function returns. It seems to me that the only workable solution would be to have a dynamic analyzer which can give you information about the types from runtime so that you can use them later when not running.

e.g. I can do the following in an interactive type window:

>>> x = etree.XPath("Hello")
>>> type(x)
<class 'lxml.etree.XPath'> 

which tells me enough that I might be able to write a stub for the "etree.XPath" function, but at best I would just make a class definition for XPath which says something like

class XPath:
pass

because I still know next to nothing about the XPath class.
Maybe I could even go so far as to type

>>> [y[0] for y in inspect.getmembers(x)]
['__call__', '__class__', '__delattr__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__le__', '__lt__', '__ne__', '__new__', '__pyx_vtable__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'error_log', 'evaluate', 'path']

which tells me that an XPath object has a member called 'evaluate' and a member called path.

I know from experience that the XPath object can be used as though it is a function itself, e.g. the docs at http://lxml.de/xpathxslt.html#the-xpath-class give the example

>>> find = etree.XPath("//b")
>>> print(find(root)[0].tag) 

where find(root) is presumably equivalent to calling find.evaluate(root). But I would have no idea how to express the idea of a "object that acts like a function but is actually a class instance" in a pyi file with my limited knowledge of type annotations.

I tried looking in my site-packages/lxml directory to see if any of the source made sense to but I was completely bewildered where to start. When I do "from lxml import etree" it must be importing etree.pyd, which is binary, which is completely opaque

[gvanrossum]

Yeah, the lxml package is touch to generate types for. Someone may have to make a real project from it... Note that stubgen is often good enough even though it doesn't generate types -- it still generates classes and methods with the right number of arguments, and the types default to Any. This still catches a lot of issues. Also, stubgen only does one module at a time. Assuming lxml has several submodules you may have to call it separately for each submodule and stitch the results together by hand. stubgen does have a way to introspect C extensions (e.g. try "python3 -m mypy.stubgen sys" and see the output in one/sys.pyi -- that's just from introspecting the module). But I don't know if there are limitations that prevent it to work on etree.pyd or not.