At 12:50 AM 5/5/2009 +0200, Tarek Ziadé wrote:

On Mon, May 4, 2009 at 7:51 PM, P.J. Eby <pje@telecommunity.com> wrote:

...

At 06:01 PM 5/4/2009 +0200, Tarek Ziadé wrote:

...

On Mon, May 4, 2009 at 5:48 PM, P.J. Eby <pje@telecommunity.com> wrote:

...

...

...

I don't see any point to the normalization.

To avoid different naming conventions like:

PKG-INFO, requires.txt, SOURCES.txt

And the problem with that is...?

inconsistency, but right, it makes no sense if any file/dir can be added there.

What about SOURCES.txt btw ? What is the reason to add it ?

It's for source distributions. It allows them to be able to rebuild an identical source distribution in the absence of source control metadata.

It's not really necessary for the installation process, although it's used to figure out which files to install if you use include_package_data=True.

Any particular reason to call it "SOURCES.txt" ?

Or we can call it MANIFEST (with '/'-separated relative path)

I called it SOURCES.txt because MANIFEST is ambiguous as to what it's a manifest *of*, and also to distinguish it from any user-generated MANIFEST file. (If you don't know about those, you're probably going to break backward compatibility w/somebody, btw. Distutils usage is a diverse collection of nightmares.) Anyway, the name of the file has no bearing on what distutils does, unless distutils is trying to implement the same feature as setuptools: i.e. round-trippable sdists whose original manifest was generated via revision control.