Consider a patch with two hunks. Apply this to a file where the hunks do not
match at the original offset but do match at a positive offset, there is a
second match for the second hunk at a negative offset.
The first hunk matches at the positive offset, the second hunk fails to match at
the original offset and the match at the negative offset is found. This
effectively reorders the hunks and the result is output that spuriously applies
the second hunk at the wrong location. This affects 1.8 and trunk.
Here is the patch:
Index: wc/f
===================================================================
--- wc/f (revision 1)
+++ wc/f (revision 2)
@@ -899,6 +899,7 @@
1898
1899
1900
+XXXX
1901
1902
1903
@@ -919,6 +920,7 @@
1918
1919
1920
+YYYY
1921
1922
1923
This is the result of applying, note the second hunk YYYY has not been output at
either the first or second match and arbitrary lines are deleted.
Index: wc/f
===================================================================
--- wc/f (revision 3)
+++ wc/f (working copy)
@@ -1211,15 +1211,17 @@
1898
1899
1900
+XXXX
1901
1902
1903
-1904
-1905
-1906
-1907
-1908
-1909
+1918
+1919
+1920
+YYYY
+1921
+1922
+1923
1910
1911
1912
Here is the script to reproduce:
#!/bin/bash -e
svnadmin create repo
svn co file://`pwd`/repo wc
for i in `seq 1000 2000`; do echo $i >> wc/f ; done
svn add wc/f
svn ci -mm wc
> wc/f
for i in `seq 1000 1900`; do echo $i >> wc/f ; done
echo XXXX >> wc/f
for i in `seq 1901 1920`; do echo $i >> wc/f ; done
echo YYYY >> wc/f
for i in `seq 1921 2000`; do echo $i >> wc/f ; done
svn ci -mm wc
> wc/f
for i in `seq 1000 1100`; do echo $i >> wc/f ; done
for i in `seq 1915 1925`; do echo $i >> wc/f ; done
for i in `seq 1101 1400`; do echo $i >> wc/f ; done
for i in `seq 0 300`; do echo ZZZZ >> wc/f ; done
for i in `seq 1401 2000`; do echo $i >> wc/f ; done
svn ci -mm wc
svn diff -c2 wc > y.y
svn patch y.y wc --strip 1
Bug
Major