Re: Meld 1.7.1 released

[Kai Willadsen <kai willadsen gmail com>]

On 3 April 2013 08:16, Angel Ezquerra <angel ezquerra gmail com> wrote:
> On Tue, Apr 2, 2013 at 10:10 PM, Kai Willadsen <kai willadsen gmail com> wrote:
> > On 3 April 2013 06:06, Angel Ezquerra <angel ezquerra gmail com> wrote:
> > > On Tue, Apr 2, 2013 at 9:50 PM, Kai Willadsen <kai willadsen gmail com> wrote:
> > > > On 3 April 2013 02:01, Keegan Witt <keeganwitt gmail com> wrote:
> > > > > I exceeded my Dropbox public folder limits, the new home for them is here:
> > > > > https://meld-installer.googlecode.com/files/meld-0.0.0.0.exe
> > > > > https://meld-installer.googlecode.com/files/meld-0.0.0.0.zip
> > > > > I'll keep these around permanently and periodically update them with the
> > > > > latest from Git master.  I include the .git directory, so you can do a git
> > > > > pull on the meld directory at any time if you want to update it yourself.
> > > > >
> > > > > This seems like an important issue to get fixed.  Do you think it'd be
> > > > > appropriate to push out another release sooner rather than later?
> > > >
> > > > Yeah, could do. I'll see if I can find time this weekend for it. For
> > > > that to happen, I'd appreciate any testing that people might want to
> > > > throw at current git. Several Windows-related fixes went in over the
> > > > weekend for Git actions, temp files and executable locating, and it
> > > > would be nice if they worked for someone other than me.
> > > >
> > > > cheers,
> > > > Kai
> > >
> > > There is a serious issue on Windows that we've been discussing on the
> > > mercurial-devel mailing list, which is that meld returns 0 even if you
> > > do not save the merge output (i.e. if the merge fails). This makes
> > > mercurial believe that the merge was successful when it isn't.
> > >
> > > For now we are thinking of enabling mercurial's "checkchanged" option
> > > for meld, which means that mercurial will check if the merge output
> > > changed once meld exits, and ask the user if the merge failed if it
> > > did not. That is a band-aid though, so improvements on that area would
> > > be really great.
> > >
> > > I think Keegan is aware of that issue. Something related to portable python?
> >
> > Not a portable python problem I think. We don't ever (to my
> > recollection) change return values. I'm not sure that it makes sense
> > to do so either, since there's no way to indicate multiple returns
> > (i.e., tabs 1 and 3 were saved but tab 2 wasn't). I realise that
> > launching multiple merges isn't a common case, but... it just feels
> > wrong to me.
> >
> > Could you explain why you consider the checkchanged option to be a band-aid?
> >
> > cheers,
> > Kai
>
> It is a band-aid because we only use that option when the tool is not
> able to tell us that the merge was successful. Most merge tools do
> tell us when a merge succeeded by returning a 0 exit code, and when
> the merge fails with a non zero exit code. This allows us to do the
> merge pretty much automatically. If the merge tool does not return a
> non zero exit code when the merge is aborted by the user then we can
> only guess, and we do so by checking if the merge file contents
> changed, which is not very accurate. If they did not change we must
> ask the user to confirm that the merge was successful, which adds an
> additional prompt.
>
> So as you see it is less than ideal. One of the main mercurial devs,
> while discussing the need to use the checkchanged setting said "That's
> a very serious issue", and I agree with him.

What I don't understand is how you expect *Meld* to know whether the
merge was successful? I mean, it's pretty easy to exit(1) if the user
doesn't save the merge file, but I don't see how that's significantly
more accurate than checking the file contents on the Mercurial side?

As I see it, the only other option is to require the user to click a
"Yep, I resolved this" button, which is functionally the same as a
command-line prompt saying "How did the merge go?".

> BTW, if I understood you correctly, this is non windows specific, right?

I haven't actually checked behaviour, but yes. There is no sensible
way right now for us to set a return code based on a single
comparison, simply because a window can contain multiple comparisons,
and we would have no idea what to return.

> I think that meld could handle the situation that you mention by
> returning 0 if the user saves any of the files, and non zero
> otherwise.

That sounds exceedingly dangerous. If you open up two merges and only
save one of them, then you're giving back a false positive, which is
much scarier than saying that a merge failed when it didn't really.

> Another option would be to only return 0 if the output file
> (as specified with the -o parameter) is saved.

Same as above I think.

> What do you think?

I'm certainly not averse to making changes to make Meld more usable by
Mercurial, but it has to not break current Meld use. At this stage, it
sounds to me like Mercurial (and maybe other systems? I don't know)
would like Meld to have a completely different mode from the normal
user-invoked one. Off the top of my head:

 * Disallow creation of new tabs
 * Disallow re-use of an existing instance
 * Add exit code based on 'success of merge' (whatever that means)
 * Some UI to indicate merge success?
 * Remove file selectors, replacing them with labels

...all of which is doable. It's just work. If there's anyone lurking
who is involved in mergetool configurations for Git or Bazaar or
whatever, now would be a good time to speak up.

cheers,
Kai