[reportlab-users] ReportLab 3.x packaging and deployment

[Marius Gedminas]

On Tue, Jan 14, 2014 at 11:19:01AM +0000, Andy Robinson wrote:

> On 14 January 2014 11:00, Marius Gedminas <marius at gedmin.as> wrote:

> > It would be awesome if I could sudo apt-get install python-reportlab and

> > then somehow make it available in selected virtualenvs, but that's a

> > discussion for another list.

> 

> Dumb questions but can't you...

> 

> 1. sudo apt-get install python-reportlab

> 2. create a virtualenv with no site packages

> 3. make a symlink from the venv's site-packages to the system reportlab?


Yes.

But I generally find

  sudo apt-get build-dep python-reportlab

and then building form source simpler.  (Because I have build scripts
that create the virtualenv and pip install -r requirements.txt for me,
and I don't want to put hacks in the middle of those scripts, and I have
no way to pause the script so I could run a custom command after
virtualenv creation but before pip install tries to pull in ReportLab
via some dependency chain.)


> I concluded long ago that pip can't do everything.  All of our web

> projects (and we have a LOT) have a shell script at the top level to

> build/rebuild the environment.   So we can check if a symlink exists

> and create one if it's missing, and do other "procedural" stuff.


I tend to have Makefiles to build virtualenvs on demand etc.  It helps
that I never have to deploy on Windows.


> A requirements.txt which specifies all dependencies is a wonderful

> thing to have anyway.  And pip is happy enough for pure when you add a

> couple more pure Python packages to tour requirements in an existing

> project overnight.  But it needs a lot of 'nursing through'.   Right

> now I'm working on someone else's financial app needing matplotlib,

> scipy (hence fortran), tables (hence HDF5 which absolutely demands a

> local source build inside the venv) etc etc.


Numpy/scipy are a bit special, but sudo apt-get build-dep helps a lot.

I think people (especially scientists who need things like Numpy/Scipy)
are recommending Conda as a solution for prebuilt binary packages that
match your platform exactly.  I haven't had a chance to try it yet.

Marius Gedminas
-- 
Q:      What's tiny and yellow and very, very, dangerous?
A:      A canary with the super-user password.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 190 bytes
Desc: Digital signature
Url : <http://two.pairlist.net/pipermail/reportlab-users/attachments/20140114/4258be81/attachment.pgp>