e-mail about case colisions and revert

gistfile1.txt

From: "Harvey Chapman" <[email protected]>
Subject: revert and remove, filename case collision behavior

Summary: 

1. Is "hg revert -a" supposed to give errors for filename case collision? I discovered that since it doesn’t, it makes for a nice way to clean up collisions on case-insensitive filesystems.
2. Also, hg remove acts a little odd when asked to remove a case collision filename. Examples below.

[ There would be an email here but the list server thinks it’s spam no matter how I send it and from where ]

I created a repo on a linux machine with two case conflicting filenames:

hg init test
cd test
touch file FILE apple
hg ci -Am “one apple and two files: file and FILE”
hg serve

I cloned it on OSX 10.10.1 using mercurial 3.2.2, and after the update failed as expected, I used the revert method mentioned in http://mercurial.selenic.com/wiki/ManualCheckout. I was surprised that revert doesn’t care about the collisions.

$ hg clone -U test_url mixed_case
$ cd mixed_case
$ hg manifest -r tip
FILE
file
$ hg update
abort: case-folding collision between file and FILE
$ hg identify
000000000000
$ hg revert -a -r tip
adding FILE
adding apple
adding file
$ hg debugrebuilddirstate -r tip
$ hg status -A
C FILE
C apple
C file
$ ls
apple file


The remove command seems to remove files using the filenames as a sort of pattern matching rather than an actual filename. Are these results correct?

Case 1: remove FILE
Result: removes FILE, deletes file

$ hg rm FILE
$ hg status -A
R FILE
! file
C apple

Case 2: remove file
Result: removes FILE and file, deletes file

$ hg rm file
removing FILE
$ hg status -A
R FILE
R file
C apple

Case 3: tried removing various mixed case spellings of file
Results: inconsistent

Remove		Marked for Removal		Extra output
1 fiLE			file
2 FIle			file					not removing FIle: file is untracked
3 fiLe			FILE				not removing fiLe: file is untracked
4 fiLE			FILE				not removing fiLE: file is untracked

Note: 1 & 4 are the same pattern, but resulted in different removals

The only consistent method I’ve found for removing the files is this:

hg rm -Af -I file

Remove		Marked for Removal
file			file
FILE		FILE
fiLE
FIle
fiLe
fiLE

If it helps others, this is how I cleaned up my real repo that had this problem where all of the duplicate files had identical content. I decided that I wanted to prefer uppercase names and used the following to remove the lowercase duplicates:

# sort manifest, ignoring case, in reverse (lowercase first),
# returning only duplicate names, again ignoring case
hg rm `hg manifest | sort -fr | uniq -id`