Re: [gentoo-dev] [v1.0.2] GLEP 74: Full-tree verification using Manifest files

[Michał Górny]

Hi,

Ok, so before we get into this deeper, here's another option we've been
discussing. Let's drop the non-strict mode entirely, drop OPTIONAL
and keep MISC as deprecated-used-to-have-special-meaning alias to DATA.

This is going to make a lot of things simpler, and avoid having the very
long discussion on what should be MISC and what not. Especially given
that the specific definition of MISC makes little sense as-is.


Two reasons have been mentioned for having non-strict mode:

1. Stripping some of non-strictly necessary files to reduce repository
size. However:

1a. Stripping of files that we can mark MISC is not going to do much.
Most of the time, people would strip whole categories or other data we
can't really mark MISC, so they will need a different solution anyway.

1b. That's just an argument for allowing them to be missing. There's no
clear reason why they would have different content, and it doesn't have
much sense to allow it implicitly.

1c. Those files can still be means of doing some kind of attacks --
starting with misinformation resulting in the user reducing security of
his systems, ending with attacks e.g. exploiting XML parser
vulnerabilities.

2. Allowing different content for cache-class files that can be updated
on user's end (e.g. md5-cache, use.local.desc...). However:

2a. We can't really do this for md5-cache since it clearly can be
abused.

2b. Again, it makes little sense since we took special care that all
those tools have stable output.


All that said, if we really have a problem that needs solving here, I'm
not convinced MISC is the right solution for it. If people need to
explicitly exclude stuff, then I suppose the configuration-injected
ignore list is much better solution for this.

-- 
Best regards,
Michał Górny