bzr read-only errors need improvement (was Re: bzr requires write permissions to root directory of repository)
John A Meinel
john at arbash-meinel.com
Mon Feb 20 15:23:35 GMT 2006
martin f krafft wrote:
> also sprach John A Meinel <john at arbash-meinel.com> [2006.02.20.1556 +0100]:
>> We don't have to have atomicity for something like .bzrignore, but
>> it is nice to have ^C not corrupt a file.
>
> you can trap that signal, no?
Yes, but there are lots of things that could go wrong while we are
writing, and I would rather not write *part* of a ignore rule.
>
>> Though I will say that doing 'bzr revert' without having write access
>> seems a little bit funny to me.
>
> yeah, i can see that. i guess i am sort of biased because i know my
> fileset will never change, which is why i can afford -w on the
> directory.
Except if you have -w on all directories. revert won't work at all.
Because IIRC we do the same atomic replace for any working tree changes.
>
>> I realize you have taken great pains to set up all of your read
>> and write access (so .bzr/ is writeable, but $wd is not, but
>> $wd/some/path is writable, etc). I think you are being a little
>> abusive to the system, but it is likely that we will be able to
>> support your use case without too much difficulty. More by
>> happenstance than by design.
>
> :)
>
>> You still have the issue of the temporary log message on commit.
>> Which we have discussed putting into .bzr/ though we really want
>> to avoid putting files a user would modify in there. I believe
>> both svn and cvs have the property that their meta directories are
>> not generally to be edited by users. (You can if you know what you
>> are doing, but bzr is the same way.) It was very bad in Arch to
>> not know which ones were okay to modify and which ones weren't.
>
> Yeah, these are very valid points. I guess I'd be very happy if bzr
> would fallback to $TMPDIR if it finds out it cannot write to .
Well, if we fix the 'last log message gets destroyed' bug, probably it
will mean saving the last log to something like .bzr/commit-log, which
means we would start and editor using that path. We would just not
recommend people edit it directly, just that it exists there, and will
be re-used if it is found. There is a subtle but important difference to
"I will start an editor with a file at this path" and "edit this file in
this path".
And we can't use TMPDIR for something like "bzr revert" or "bzr merge",
because of the atomic rename stuff. TMP is not guaranteed to be on the
same filesystem.
John
=:->
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 249 bytes
Desc: OpenPGP digital signature
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20060220/58a45a59/attachment.pgp
More information about the bazaar
mailing list