Re: [gentoo-dev] [PATCH 01/14] cdrom.eclass: Detect case-insensitively and handle special characters

[James Le Cuirot <chewi@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 2933 bytes --]

On Wed, 19 Apr 2017 05:14:34 +0200
Michał Górny <mgorny@gentoo.org> wrote:

> >> > @@ -243,4 +240,27 @@ _cdrom_locate_file_on_cd() {
> >> >  	done
> >> >  }
> >> >  
> >> > +# @FUNCTION: _cdrom_glob_match
> >> > +# @USAGE: <root directory> <path>
> >> > +# @INTERNAL
> >> > +# @DESCRIPTION:
> >> > +# Locates the given path ($2) within the given root directory ($1)
> >> > +# case-insensitively and returns the first actual matching path.  
> >This  
> >> > +# eclass previously used "find -iname" but it only checked the  
> >file  
> >> > +# case-insensitively and not the directories. There is "find  
> >-ipath" but  
> >> > +# this does not intelligently skip non-matching paths, making it
> >> > +# slow. Case-insensitive matching can only be applied to patterns  
> >so  
> >> > +# extended globbing is used to turn regular strings into patterns.  
> >All  
> >> > +# special characters are escaped so don't worry about breaking  
> >this. The  
> >> > +# first person to make this work without an eval wins a cookie.
> >> > +_cdrom_glob_match() {
> >> > +	local p=\?\($(sed -e 's:[^A-Za-z0-9/]:\\\0:g' -e 's:/:)/?(:g' <<<  
> >"$2" || die)\)    
> >> 
> >> Explanatory comment needed, i.e. what gets converted into what, and  
> >why.
> >
> >I'll add this:
> >
> ># The following line turns this:
> >#  foo*foo/bar bar/baz/file.zip
> >#
> ># Into this:
> >#  ?(foo\*foo)/?(bar\ bar)/?(baz)/?(file\.zip)
> >#
> ># This turns every path component into an escaped extended glob
> ># pattern to allow case-insensitive matching. Globs cannot span
> ># directories so each component becomes an individual pattern.  
> 
> Why do you escape pattern chars? Wasn't the variable supposed to be a
> pattern in the first place?

If you mean in the eclass before I changed it then no. In the non
CD_ROOT case, the basename was passed to "find -iname" but this was not
documented. In the CD_ROOT case, the whole thing was passed to
[[ -e ]] so patterns wouldn't have worked here.

You wouldn't want to use a pattern anyway as you're trying to uniquely
identify the disc using a very specific filename. Conversely, I relaxed
case-sensitivity because this can vary depending on whether we're
dealing with the original disc, a copy of some kind, or an existing
installation that may come from Windows.

The Curse of Monkey Island turned out to be a great example. Both discs
have some files in common like COMI.LA0, however, when mounted with the
default options, it appears upper-cased on the first disc but
lower-cased on the second. Why? The second disc doesn't use Joliet as
all the filenames have the old 8.3 format. Linux normalises these to
lower-case. The first disc does use Joliet because of a single file,
"Curse of Monkey Island - Manual.pdf" so all the other 8.3 filename are
left as upper-case by Linux.

-- 
James Le Cuirot (chewi)
Gentoo Linux Developer

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 981 bytes --]