Re: [PATCH] gitignore: add *.bz2 and *.cpio to top-level; clean up usr/

["H. Peter Anvin" <hpa@zytor.com>]

Sam Ravnborg wrote:
> 
> My concern is that we may decide to carry files in certain formats
> in the kernel source.
> And I see a tendency to add more and more file extensions to the
> top-level .gitignore file.
> 
> It is fine as long as this is files that are:
> 1) either generated in a lot of places
> 2) or generated in the top-level directory
> 
> But files that we generate in a few arch/*/boot/ directories
> does not belong in the top-level .gitignore file.
> We should keep the ignore rules close to where they apply,
> even if this may cause us to add a few more lines
> to the relevant .gitignore files.
> 

Honestly, I think this is ridiculous.  A single well-maintained
.gitignore file is a *service* to the whole tree, and the last thing we
want is git to behave differently in different subdirectories.

It is *much better* to have global rules, and add exceptions out in the
leaves of the tree where they apply.  The question that the global
.gitignore should answer is:

"If I have a file of type X, is the user *likely* to want to actually
want it in the tree?"

In the case of *.gz *.bz2 *.lzma or *.cpio, I think the answer is a
resounding "no".  Almost every architecture uses compressed files at
some stage of its boot, and it's *always* a generated file.  A
non-generated file is probably a patch being handled by a developer, not
something that is meant to be in the tree.

I'm not saying we wouldn't *ever* want to have compressed binary blobs
in the tree -- there are actually a handful of reasons to do so -- but
*those are the exceptions*, and therefore should be using !-rules in
their respective .gitignores, instead of requiring that the entire tree
suffers from less sensible defaults.

	-hpa