DIFF(1)


NAME   diff - Public Domain diff (context diff) program


SYNOPSIS   diff [-b -c -i -e] file1 file2


DESCRIPTION

Diff compares two files, showing what must be changed to make them
identical.  Either file1 or file2 (but not both) may refer to
directories.  If that is the case, a file in the directory whose
name is the same as the other file argument will be used.  The
standard input may be used for one of the files by replacing the
argument by "-".  Except for the standard input, both files must
be on disk devices.


OPTIONS

    -b   Remove trailing whitespace (blanks and tabs) and compress
         all other strings of whitespace to a single blank.

    -c   Print some context -- matching lines before and after the
         non-match section.  Mark non-matched sections with "|".

    -i   Ignore lower/upper case distinctions.

    -e   Output is in an "editor script" format which is compatible
         with the Unix 'ed' editor.


All information needed to compare the files is maintained in main
memory.  This means that very large files (or fairly large files
with many differences) will cause the program to abort with an
"out of space" message.  Main memory requirements (in words) are
approximately:

      2 * (length of file1 + length of file2) + (3 * number of changes)

where "length" is the number of lines of data in each file.


The algorithm reads each file twice, once to build hash tables
and once to check for fortuitous matches (two lines that are in
fact different, but which have the same hash value).  CPU time
requirements include sorting the hash tables and randomly
searching memory tables for equivalence classes.  For example, on
a time-shared VAX-11/780, comparing two 1000 line files required
about 30 seconds (elapsed clock time) and about 10,000 bytes of
working storage.  About 90 per-cent of the time was taken up by
file I/O.


DIAGNOSTICS

Warning, bad option 'x':   The option is ignored.

Usage ...:   Two input files were not specified.

Can't open input file "filename":   Can't continue.

Out of space:   The program ran out of memory while comparing the
                two files.

Can't read line nnn at xxx in file[A/B]:   This indicates an I/O error
                                           when seeking to the specific
                                           line.  It should not happen.

Spurious match, output is not optimal:   Two lines that were different
                                         yielded the same hash value.
                                         This is harmless except that the
                                         difference output is not the
                                         minimum set of differences between
                                         the two files.  For example, instead
                                         of the output:

    lines 1 to 5 were changed to ...     the program will print

    lines 1 to 3 were changed to ...
    lines 4 to 5 were changed to ...
                                         The program uses a CRC16 hash code.
                                         The likelihood of this error is
                                         quite small.


AUTHOR

The diff algorithm was developed by J. W. Hunt and M. D. McIlroy,
using a central algorithm defined by H. S. Stone.

It was published in:

    Hunt, J. W., and McIlroy, M. D.,
    An Algorithm for Differential File Comparison,
    Computing Science Technical Report #41,
    Bell Laboratories, Murray Hill, NJ  07974


BUGS

On RSX and DECUS C on VMS systems, diff may fail if the both
files are not "variable-length, implied carriage control" format.
The scopy program can be used to convert files to this format if
problems arise.

When compiled under VAX C, diff handles STREAM_LF files properly
(in addition to the canonical variable-length implied carriage
control files).  Other variations should work, but have not been
tested.


When compiled under VAX C, diff is quite slow for unknown reasons
which ought to be investigated.  On the other hand, it has access
to effectively unlimited memory.


Output in a form suitable for ed - the -e option - seems rather
pointless; the analogue on DEC systems is SLP (SUMSLP on VMS).  It
would be simple to provide SLP-compatible output.  The question
is, why bother - since the various DEC file comparison utilities
already produce it.

============================================================================

README:

17-JAN-88

This is a enhanced version of the diff program that was posted to
comp.sources.misc recently. The diff now uses less memory, is almost
as fast as the UNIX diff and it produces "new style" context diffs.

                                Johan Widen
                                jw@sics.se

============================================================================

README.old:

Here's a public domain diff with the -b and -c options.  (4.2bsd style
contex diffs.)  I wasn't aware that these wern't present in all UNIX
versions of diff, so I didn't think posting it was a priority.

It's large, slow, and many of the comments are no longer true, but it
does work  (except when it runs out of memory).  The one case I know of
where its output is incompatable with patch does seem to be pretty
rare.

No makefile is included, the 4.2bsd diff is better on the unix system I
If you don't know how to compile and load a single C program, this probably
isn't the tool for you anyway.

I'd be grateful to anyone who cleans this up and documents it properly.
It does appear to have been separate files at some point; I'm presenting
it in a form similar to how I got it:  mail headers and outdated
documentation in comments and all.  I just banged on it enough to get it
doing what I wanted.