[PATCH] filemerge: more backwards compatible behavior for ui.merge

[Matt Mackall]

On Mon, 2008-02-04 at 22:27 -0600, Steve Borho wrote:
> On Mon, 2008-02-04 at 22:24 -0600, Matt Mackall wrote:
> > On Mon, 2008-02-04 at 21:55 -0600, Steve Borho wrote:
> > > On Mon, 2008-02-04 at 21:47 -0600, Steve Borho wrote:
> > > > # HG changeset patch
> > > > # User Steve Borho <steve at borho.org>
> > > > # Date 1202183245 21600
> > > > # Node ID ce3bcf0fd12054659f7798c46cea4554618ae1b5
> > > > # Parent  532093498c8f7953d01a1bc3d874deee24d8be24
> > > > filemerge: more backwards compatible behavior for ui.merge
> > > > 
> > > > ui.merge is parsed into (tool, args).  [merge-tools] is still referenced
> > > > to find tool.properties, but if args are specified in ui.merge they take
> > > > precedence over merge-tools.tool.args.
> > > This patch is flawed.  It shouldn't just string join the args back
> > > together because it loses the original grouping in ui.merge.
> > > 
> > > I'm mainly curious whether this is preferable behavior for ui.merge.
> > > If this is an acceptable approach, I'll submit a fixed patch.
> > 
> > Not sure about this at all. I think we basically want two cases:
> > 
> > a) ui.merge = <tool> -> use [merge-tool] settings
> > b) ui.merge = <tool-with-args> -> don't
> > 
> > And I think that just works now, no?
> 
> Option b) fails find_exe() unless the command line has os.sep in it, as
> asak pointed out.

Thought your earlier patch fixed that? Ahh, I guess not.
-- 
Mathematics is the supreme nostalgia of our time.