Next: , Previous: Adjusting Output, Up: Top

6 diff Performance Tradeoffs

GNU diff runs quite efficiently; however, in some circumstances you can cause it to run faster or produce a more compact set of changes.

One way to improve diff performance is to use hard or symbolic links to files instead of copies. This improves performance because diff normally does not need to read two hard or symbolic links to the same file, since their contents must be identical. For example, suppose you copy a large directory hierarchy, make a few changes to the copy, and then often use ‘diff -r’ to compare the original to the copy. If the original files are read-only, you can greatly improve performance by creating the copy using hard or symbolic links (e.g., with GNUcp -lR’ or ‘cp -sR’). Before editing a file in the copy for the first time, you should break the link and replace it with a regular copy.

You can also affect the performance of GNU diff by giving it options that change the way it compares files. Performance has more than one dimension. These options improve one aspect of performance at the cost of another, or they improve performance in some cases while hurting it in others.

The way that GNU diff determines which lines have changed always comes up with a near-minimal set of differences. Usually it is good enough for practical purposes. If the diff output is large, you might want diff to use a modified algorithm that sometimes produces a smaller set of differences. The --minimal (-d) option does this; however, it can also cause diff to run more slowly than usual, so it is not the default behavior.

When the files you are comparing are large and have small groups of changes scattered throughout them, you can use the --speed-large-files option to make a different modification to the algorithm that diff uses. If the input files have a constant small density of changes, this option speeds up the comparisons without changing the output. If not, diff might produce a larger set of differences; however, the output will still be correct.

Normally diff discards the prefix and suffix that is common to both files before it attempts to find a minimal set of differences. This makes diff run faster, but occasionally it may produce non-minimal output. The --horizon-lines=lines option prevents diff from discarding the last lines lines of the prefix and the first lines lines of the suffix. This gives diff further opportunities to find a minimal output.

Suppose a run of changed lines includes a sequence of lines at one end and there is an identical sequence of lines just outside the other end. The diff command is free to choose which identical sequence is included in the hunk. In this case, diff normally shifts the hunk's boundaries when this merges adjacent hunks, or shifts a hunk's lines towards the end of the file. Merging hunks can make the output look nicer in some cases.