Dealing with case insensitivity

[Bryce McKinlay <bryce at waitaki dot otago dot ac dot nz>]

I wonder if anyone has any ideas on what is the best way to change jc1 
so that it can tolerate case insensitive file systems.

On Darwin, the libjava build dies immediately with:

/Users/bryce/cvs/gcc/build/gcc/gcj 
-B/Users/bryce/cvs/gcc/build/powerpc-apple-darwin1.4/libjava/ 
-B/Users/bryce/cvs/gcc/build/gcc/ --encoding=UTF-8 -C -g -classpath 
/Users/bryce/cvs/gcc/build/powerpc-apple-
darwin1.4/libjava:../../../libjava -d 
/Users/bryce/cvs/gcc/build/powerpc-apple-darwin1.4/libjava 
java/lang/ConcreteProcess.java
/Users/bryce/cvs/gcc/libjava/gnu/java/security/provider/gnu.java:31: 
Class `gnu.java.security.provider.Gnu' already defined in 
/Users/bryce/cvs/gcc/libjava/gnu/java/security/provider/Gnu.java:31.
    public final class Gnu extends Provider
                       ^

At some point the compiler tries to load "gnu.java" (because "gnu" is a 
package name?), which succeeds unexpectedly, and the compiler thinks 
there are two different files defining the same class.

There was a patch posted some time ago which forced everything to 
lower-case, but that wouldn't be right for case-sensitive systems, and 
doesn't seem to fully work anyway.

I tried to get it to canonicalise all path names using realpath() before 
using BUILD_FILENAME_IDENTIFIER_NODE in jcf-parse, but realpath didn't 
do what I want - it just returns the filename in the case given to it.

One solution might be to stat() all filenames before we open them, and 
keep the file tree nodes in a hashtable indexed by their inode number. 
This way we can be sure that the file nodes are unique. Any problems 
with that approach? Anyone see better way?

regards

Bryce.