Bug 449850

Description of problem:
A diff between two identical files (one using DOS newlines, the other UNIX
newlines) using the -w flag to ignore the whitespace differences takes an
inordinately long time.

Version-Release number of selected component (if applicable):
Bug observed on a number of Redhat platforms.

How reproducible:
very high

Steps to Reproduce:
1. Create a text file. A high number of lines is better than longer lines.
2. Convert to/from DOS/UNIX newlines
3. `diff -w $FILE_1 $FILE_2`
  
Actual results:

The output of `time diff -w $DOS_FILE $UNIX_FILE` with the Redhat supplied diff was:
real    6m21.679s
user    3m54.609s
sys     0m0.117s


Expected results:

The output of `time diff -w $DOS_FILE $UNIX_FILE` with GNU diff from source was:
real    0m5.954s
user    0m0.179s
sys     0m0.037s


Additional info:

The expected results have been corroborated on Debian 4, Ubuntu 8, Solaris SPARC
8, 9, Mac OS X 10.5.2, and FreeBSD 6.2 using their GNU diff, or their
vendor-specific diff.

This behavior has been replicated in RHEL 3, 4, 5.1, CentOS 4.5, Mandriva 2008,
and OpenSUSE 10. Has not been tried in Fedora Core 9, but the existing pattern
leads me to believe it will exist.

This bug appears to work on a geometric scale as a 50MB file of similar format
to the supplied example files will take no less than ~10 hours to diff on a
computer faster than 2.0GHz. During this time the diff process will max a single
CPU/HT/Core but occupy no more than the two files and working space in RAM.

There appears to be no performance difference between physical hardware or a
virtualized environment.