Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Setuptools packaging improvments #1647

Closed
wants to merge 10 commits into from

Conversation

dennybaa
Copy link
Contributor

Setup.py is not used and not working now as it should. Basically, we devs, just need pip, virtualenv and requirements.txt. However it's interesting that setup.py is really useful when it comes to packaging and distribution.

I've started to work on clearing up st2 dependency and packaging problems. There are a few ideas I suggest to take into work.

  1. We keep on using requirements.txt and *-requirements.txt as the source of dependency truth.
  2. We must include dependencies into the specific component requirements.txt file, as we decided to distribute each component separately (this is consequence of our choice).
  3. We keep using all-in-one requirements.txt and *-requirements.txt for the current workflow compatibility. But basically it should be kept, imo, just because it's handy to setup all deps for st2 on dev machine...

As the result and after proper setup.py fixes, component packages should be able to be built and installed with setup.py. This will allow us packaging and distributing st2 components using pypi as well as debs, rpms etc since majority of tools rely on the python native setup.py method.

PS. I'd better use naming dev-requirements.txt rather than test-*, imo.

@Kami
Copy link
Member

Kami commented Jun 20, 2015

I will have a look shortly.

In general I agree that we need a working setup.py, especially since we eventually plan do distribute components as wheels.

from setuptools import setup, find_packages


def fetch_requirements():
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

pip.req.parse_requirements should work as long as a new version of pip is available on the system.

See tweepy/tweepy#533 for an example.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Edit: Nvm, I didn't see the full diff. You already use that function :)

On a related note - we just need to make sure we use pip >= 6.0 since older versions don't take session argument.

@Kami
Copy link
Member

Kami commented Jun 20, 2015

If we keep and have two requirements file which are managed manually (global and one per component), things will be hard to maintain and get out of sync.

We can update Makefile to install dependencies from each component and maybe also add a new target which combines all the requirements in a global requirements file.

@Kami
Copy link
Member

Kami commented Jun 20, 2015

Overall, nice work 👍

We also need to make sure we add a new integration test which tries to install each component (e.g. python setup.py install and / or using pip), otherwise I'm pretty sure things will get out of sync and break soon.


current_dir = os.path.dirname(os.path.realpath(__file__))
version_file = os.path.join(current_dir, '../st2client/st2client/__init__.py')
with open(version_file, 'r') as f:
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To make this a bit more robust - can we just temporary add st2client package to the path and then do something like:

from st2client import __version__
version = __version__.split('.')

This should work as long as init.py in st2client doesn't depend on other external dependencies (which is a bad practice anyway).

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry but we can't use it as well as my method with parsing file. This limit is due to our potential tooling when wheel is created (or egg) - pip is run and component directory copied under some '/tmp/$RANDOM$' path.
That's why we can't use relative file read as well as your approach.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've added top st2_version file, which is being populated for each component also if it can not be found. This is done by make, so by the type setup.py is run, all is nice file can be cat.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

IIRC, dict comprehensions are only available in 2.7 and above and we might
still run tests with 2.6.
On Jun 23, 2015 1:41 AM, "Denis Baryshev" [email protected] wrote:

In st2api/setup.py
#1647 (comment):

+from setuptools import setup, find_packages
+
+
+def fetch_requirements():

  • links = []
  • reqs = []
  • for req in parse_requirements('requirements.txt', session=False):
  •    if req.link:
    
  •        links.append(str(req.link))
    
  •    reqs.append(str(req.req))
    
  • return (reqs, links)

+current_dir = os.path.dirname(os.path.realpath(file))
+version_file = os.path.join(current_dir, '../st2client/st2client/init.py')
+with open(version_file, 'r') as f:

I've added top st2_version file, which is being populated for each
component also if it can be found. This is done by make, so by the type
setup.py is run, all is nice file can be cat.


Reply to this email directly or view it on GitHub
https://github.com/StackStorm/st2/pull/1647/files#r32961529.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In short - for now we still run tests and other lint tasks under Python 2.6 so the easiest / best thing to do is to just replace dict comprehension with dict([list comprehension]).

When we decide to drop support for 2.6 we can re-evaluate that.

@dennybaa
Copy link
Contributor Author

That's a nice idea to create a Makefile, so that we manage only
component specific requirement's. Right?

On Sat, Jun 20, 2015 at 12:24 PM, Tomaz Muraus
[email protected] wrote:

If we keep and have two requirements file which are managed manually
(global and one per component), things will be hard to maintain and
get out of sync.

We can update Makefile to install dependencies from each component
and maybe also add a new target which combines all the requirements
in a global requirements file.


Reply to this email directly or view it on GitHub.

@dennybaa
Copy link
Contributor Author

Thanks)

On Sat, Jun 20, 2015 at 12:26 PM, Tomaz Muraus
[email protected] wrote:

Overall, nice work

We also need to make sure we add a new integration test which tries
to install each component (e.g. python setup.py install and / or
using pip), otherwise I'm pretty sure things will get out of sync and
break soon.


Reply to this email directly or view it on GitHub.

@dennybaa
Copy link
Contributor Author

Don't actually merge so far) This is just an idea. If you like it I can keep this up to date with my work.
I've actually started this, because:

  1. I want specific requirements.txt for any component and just to keep it clear and sane.
  2. I want to build solid stand-alone debian packages (without post-inst hackary)

@@ -1,3 +1,5 @@
# 1. Don't include forget to populate spcific requirements.txt. IMORTANT!!!
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: typo. specific. IMPORTANT.
Nit: Sentence isn't making sense.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ha-ha. You are right) This sentence will likely disappear from here soon,
let's ignore it meanwhile)

On Tue, Jun 23, 2015, 07:53 Lakshmi Kannan [email protected] wrote:

In requirements.txt
#1647 (comment):

@@ -1,3 +1,5 @@
+# 1. Don't include forget to populate spcific requirements.txt. IMORTANT!!!

Nit: typo. specific. IMPORTANT.
Nit: Sentence isn't making sense.


Reply to this email directly or view it on GitHub
https://github.com/StackStorm/st2/pull/1647/files#r33009336.

@lakshmi-kannan
Copy link
Contributor

Overall, seems to be proceeding in right direction but I am not a setup tools expert. I am learning as well.

from distutils.version import StrictVersion


cwd = os.path.abspath(os.getcwd())
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of using cwd, it would be safer to use location where this file is located and work from there.

BASE_DIR = os.path.dirname(os.path.abspath(__file__))
...

@Kami
Copy link
Member

Kami commented Jun 24, 2015

Can you please also sync your branch with latest master?

I've added some changes so we not also lint Python files in scripts/ directory.

# https://docs.python.org/2/distutils/sourcedist.html#commands
# Include all files under the source tree by default.
# Other behaviour can be used though.
recursive-include st2common *.* *
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Kami, @lakshmi-kannan there was a question why I have removed package_data, it's because the way it works we must check it carefully. Unless we are not sure what files to include better to include all sources and here comes MANIFEST.in..

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of just including all in the manifest, I would prefer to explicitly declare all the non-code artifacts we want to include here - so things like README, LICENSE, NOTICE, requirements.txt, etc.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yep, so far I suggest like this. Later on more comprehensive filtering can
be done.
BTW none code artifacts are included as you might think. Read this as
st2common/{,.*} in bash globing way.
MANIFEST.in is relative to the component root, i.e we include files not
under component root but only under its source folder.

On Thu, Jun 25, 2015, 08:37 Tomaz Muraus [email protected] wrote:

In st2common/MANIFEST.in
#1647 (comment):

@@ -0,0 +1,4 @@
+# https://docs.python.org/2/distutils/sourcedist.html#commands
+# Include all files under the source tree by default.
+# Other behaviour can be used though.
+recursive-include st2common . *

Instead of just including all in the manifest, I would prefer to
explicitly declare all the non-code artifacts we want to include here - so
things like README, LICENSE, NOTICE, requirements.txt, etc.


Reply to this email directly or view it on GitHub
https://github.com/StackStorm/st2/pull/1647/files#r33223754.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm still confused by that line - "st2common/{,.*}" basically means everything under st2common/st2common.

In any case, like I said above, I would prefer to be explicit and collect Python modules and packages we want to include inside setup.py and only explicitly declare other non-package files we want to include here.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's explicit, all stuff under st2common/st2common. That's package_data works quite implicitly...
I mean you add some files under st2common/st2common/htmls, and setup.py breaks the things since it doesn't include htmls. And so on and so forth.
Including all the source is the most explicit behavior) It's useful now and might be valid in the future.

And package_data - isn't used in 100% of cases, it's limited... and its usage prone to errors, imo. MANIFIST.in is more comprehensive way to go. It's seems not important so far, but I suggest strictly include everything under code and future will introduce new use cases.

@dennybaa
Copy link
Contributor Author

I've taken this from travis

# Merge into one st2 components-wide requirements.txt file.
python ./scripts/fixate-requirements.py -s st2*/in-requirements.txt -f fixed-requirements.txt
  File "./scripts/fixate-requirements.py", line 110
    fixedreq_hash = {req.req.project_name: req for req in fixed if req.req}
                                                 ^
SyntaxError: invalid syntax
make: *** [requirements] Error 1

Maybe I don't get smth or missing, it looks like a valid dict comprehension, moreover I've run this code. @Kami could you give a clue.

@Kami
Copy link
Member

Kami commented Jun 25, 2015

Another thing I just remembered - we should also ship tests with each component (tests/ sub-directory) and those tests should be runnable.

Most of the tests depend on st2tests package which means we also need to ship this package with each component otherwise user won't be able to run component specific tests.

@Kami
Copy link
Member

Kami commented Jun 25, 2015

Please let me know when all the comments have been addressed and I will go, check out the branch and test the changes locally.

In addition to that, I will probably also make some small changes (add a readme to each component, make sure we include license and notice file with each component, etc.).

@dennybaa
Copy link
Contributor Author

@Kami, regarding st2tests.
We should not ship tests with packages, it's not needed. This kind of tests should better stay on dev boxes and not deployed to production. It's an overkill, that's why it's not needed.
Production tests may exist though, but it's a different story from unit testing and bdd testing.

@Kami Kami changed the title [WIP] Setuptools packaging improvments Setuptools packaging improvments Jun 30, 2015
@Kami
Copy link
Member

Kami commented Jun 30, 2015

Closing in favor of #1668.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants