Backward merge causes random outputs in test-win32text

[Patrick Mézard]

Mercurial Commits a écrit :
> http://hg.intevation.org/mercurial/crew/rev/a964ab624385
> changeset:   8742:a964ab624385
> user:        Matt Mackall <mpm at selenic.com>
> date:        Mon Jun 08 18:14:44 2009 -0500
> summary:     merge: allow merging going backwards

It breaks test-win32text sometimes, both on Windows and OSX. The last update -C of the test behaves randomly. If I had a --debug:

ERROR: test-win32text output changed
--- Expected output 
+++ Test output 
@@ -314,6 +314,18 @@
 [encode]
 ** = !
 
+resolving manifests
+ overwrite True partial False
+ ancestor 46f1d1c130eb+ local 46f1d1c130eb+ remote 46f1d1c130eb
+ linefeed: remote is newer -> g
+ bin: remote created -> g
+ f3: remote created -> g
+ f4.bat: remote created -> g
+getting bin
+getting f3
+getting f4.bat
+getting linefeed
 4 files updated, 0 files merged, 0 files removed, 0 files unresolved
 hello<NUL><CR><LF>
 some<LF>text<LF>
!


ERROR: test-win32text output changed
--- Expected output 
+++ Test output 
@@ -314,7 +314,16 @@
 [encode]
 ** = !
 
-4 files updated, 0 files merged, 0 files removed, 0 files unresolved
+resolving manifests
+ overwrite True partial False
+ ancestor 46f1d1c130eb+ local 46f1d1c130eb+ remote 46f1d1c130eb
+ bin: remote created -> g
+ f3: remote created -> g
+ f4.bat: remote created -> g
+getting bin
+getting f3
+getting f4.bat
+3 files updated, 0 files merged, 0 files removed, 0 files unresolved
 hello<NUL><CR><LF>
 some<LF>text<LF>
 rem empty<LF>
!


The considered revisions are the same, but it seems linefeed is sometimes considered for updating, sometimes not.

--
Patrick Mézard