[cvsnt] Unicode bug

[Hiroshi Kuribara]

Tony Hoyle wrote:

> > Aside from binary problem, I found many bugs in -ku mode.
> > So I forbid -ku option in my office.
> > 
> It would be helpful if you said what you thought those bugs are - people use
> -ku all the time without problems... there hasn't been a bug reported
> involving it in many months.

In MY environment, that is win2k with cvsnt2.0.14

* -ku bug

1. add and commit some lines unicode file with little endian bom.
2. add, delete and change line of the above file and commit
3. retreive the past version with 'cvs update -rx.x ***'
(It's most basic action, not special)

In pserver mode, update fails with message like below

P test.txt
cvs client: refetching unpatchable files
U test.txt
cvs update: checksum failure after patch to ./test.txt; will refetch

In :local: mode, failed with below

cvs [update aborted]: invalid change text in C:/Work/CVSWORK/KuriTest/test.txt,v

I must apologize that from the first stage of test cvsnt doesn't work,
so I can say only one bug.
But my experience of this, after you fix the above one,
you'll encounter other related bugs.

:local: and client/server independently,

When you changed the HEAD file, and update for merging with other sandbox's commit.
When you changed the file that is not HEAD file, and update to HEAD with update -A.

And that might not be bug, might be specification.
1. checkout (-ku)unicode file with -kt options. You get utf8 file(I like this).
2. Edit and commit.
Repository file will be broken.
We should prevent this type actions.

By the way, my english is so bad.
Please fogive me for politeless.

/Hiroshi


__________________________________________________
Do You Yahoo!?
Yahoo! BB is Broadband by Yahoo!
http://bb.yahoo.co.jp/