X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=finddup.1;h=c86df1ea89daff89f228e2a87944407689e4ebb1;hb=2e218d38854bc5934f3935bc7aa6eab598770bdb;hp=18c27e5f57def998cccccfbcbfb1c3bc3907fe48;hpb=c54c57cbd82dd37632d9ed551ebab0c1489dfd69;p=finddup.git

diff --git a/finddup.1 b/finddup.1
index 18c27e5..c86df1e 100644
--- a/finddup.1
+++ b/finddup.1
@@ -66,17 +66,24 @@ files with same inode are considered as different
 
 None known, probably many. Valgrind does not complain though.
 
+The current algorithm is dumb, as it does not use any hashing of the
+file content.
+
+Here are the things I tried, which did not help at all: (1) Computing
+md5s on the whole files, which is not satisfactory because files are
+often never read entirely hence the md5s can not be properly computed,
+(2) computing XOR of the first 4, 16 and 256 bytes with rejection as
+soon as one does not match, (3) reading parts of the files of
+increasing sizes so that rejection could be done with a small fraction
+when possible, (4) using mmap instead of open/read.
+
 .SH "WISH LIST"
 
 The format of the output should definitely be improved. Not clear how.
 
-The comparison algorithm could maybe be improved with some MD5 kind of
-signature. However, most of the time is taken by comparison for
-matching files, which are required even when using a hash.
-
 Their could be some fancy option to link two instances of the command
-running on different machines to reduce network disk accesses. Again,
-this may not help much, for the reason given above.
+running on different machines to reduce network disk accesses. This
+may not help much though.
 
 .SH "EXAMPLES"