more performance work?
Lukáš Lalinský
lalinsky at gmail.com
Thu Nov 8 10:04:01 GMT 2007
On St, 2007-11-07 at 23:33 -0500, Martin Pool wrote:
> I've come to agree that line ending conversion should store things in
> a canonical form, rather than just presenting the checked-out files
> appropriately.
>
> I don't object to adding file properties but I'm not sure they are
> necessary to support this, rather than a configuration file saying
> that particular filenames should be treated in particular ways. I
> would think generally you'd want expansion or line mangling across a
> class of files...
Well, if the only purpose of this is converting from LF to CRLF on
checkout, and the other way around on commit, then perhaps a
configuration file would be enough.
But why not take this farther and solve other issues bzr currently has?
For example, it will currently try to diff and merge some PDF files, and
refure to diff and merge UTF-16 XML files. Or diffs for UTF-8 encoded
files will display as garbage on non-UTF-8 console, while commit
messages and author names will display just fine. For these kind of
things, I'd prefer Perforce-like per file property "type", with options:
binary (no conversions), text (stored as LF + automatic conversions),
unicode (stored as LF/UTF-8 + automatic conversions). Or we could even
use the current "kind" property for this, since these special cases will
almost always require special handling just like for example file vs.
symlink.
Lukas
More information about the bazaar
mailing list