From: Dave Thomas <Dave@...>
Date: 2002-05-07T12:38:51+09:00
Subject: Re: require and case

Marko Schulz <in6x059@public.uni-hamburg.de> writes:

> > that rb_require should normalize all paths before executing, and
> > should store those normalized paths in $".
> 
> That will likely break scripts that fiddle around with $:. So it
> probably is unwise to change it in a minor release.

Surely the only way it could break a script is that if that script
currently manipulates $: in order deliberately to require the same
file twice. Making names canonical before storing them in $" _removes_
the impact of changing $:, doesn't it?

However, I have had problems with scripts that manipulate $: loading
the same file twice: that was an issue with rubicon for a while. 

> > Thus on a Windows box the names should be converted to lower case
> > and expand_path called.
> 
> Does expand_path() solve the problem that on pre-NTFS-filesystems
> every file with a name that is too long has two different names?
>   parsedate.rb vs. parsda~1.rb

No idea - maybe it isn't expand_path, but something else that returns
the canonical name. I'll happily leave that to someone who understands
Windows.

> > On a Unix both, the case should be left unchanged but expand_path is
> > called.
> 
> expand_path() doesn't resolve symbolic links.
> 
> If there is more than one hard link to a file it is impossible to
> determine _the_ correct name. I don't know NTFS, but this might be an
> issue there too.

We could store the inode number instead.

> And there are so many special cases how a file may be addressed that
> it is virtually impossible for ruby to ensure that a file gets loaded
> only once.  So IMHO it should stick with the simple and understandable
> route it takes at the moment.

The current scheme _has_ bitten me - not because of the case of file
names, but because "./fred" is different to "fred". That surprised me.


Dave