Re: merging stable to default fails

[Daniel J Sebald]

On 01/12/2013 01:31 AM, Daniel J Sebald wrote:
> On 01/11/2013 09:27 PM, John W. Eaton wrote:
> > When trying to merge stable to default I see the following error from
> > hg:
> >
> > $ hg merge stable
> > abort: path contains illegal component: /OctJavaQry.class
> >
> > Any clues? Rik added the binary file OctJavaQry.class to the hg
> > archive, but I have no clue how that could cause the error above.
> >
> > jwe

[snip]

> Perhaps Mercurial believes
>
> store/fncache:data/build-aux/OctJavaQry.java.i
>
> is illegal because it should be
>
> store/fncache:data/build-aux/_oct_java_qry.java.i
>
> inside of "store/fncache".
>
> If this is the case, how this came about, I can only guess. Perhaps this
> is a Windows/others issue because of DOS ignoring case or something?
> Maybe the Mercurial settings on Rik's machine are different than yours?
> (guessing!) Might there be a command line switch controlling how upper
> case letters are treated in the database?

I doubt that theory.  Looking through fncache, I see plenty of file 
names with upper case letters intact.

How about the following link?  Here seems like a clue:

http://permalink.gmane.org/gmane.comp.version-control.mercurial.general/31328

If I understand correctly, the data/build-aux of

data/build-aux/OctJavaQry.class.i

is being stripped in the merge process and creating /OctJavaQry.class.i. 
 (Note the link in that post to a line of code in Mercurial that does 
this.)

However, in the link above it explains that this is due to movement of 
the file in the tree somehow.  That doesn't seem to be the scenario 
here.  I do notice something unique and peculiar about OctJavaQry.class 
which only has a couple associated change logs in the "default" branch.

http://hg.savannah.gnu.org/hgweb/octave/log/450e4e60d2c9/build-aux/OctJavaQry.class

[As an aside, I sure wish Mercurial interactive interface had just a tad 
more convenient info.  For example, follow into one of the branches and 
there is no indication anywhere what the branch is.  Just have to have a 
good memory I guess after going a few directories deep.  Also, the graph 
mode has color-coded branch indications with little dots for the where 
the associated changeset belongs.  However, there is no key for what 
color belongs to what branch.  Maybe I'm missing something.]

Anyway, here is the file log for the entry you are having problems with:

http://hg.savannah.gnu.org/hgweb/octave/log/450e4e60d2c9/build-aux/OctJavaQry.class

If one follows either of those changesets, he'll find that there is no 
particular entry for OctJavaQry.class like there is for 
/build-aux/OctJavaQry.java.  Because OctJavaQry.class is binary, it's as 
though it is treated as a Git sub-repository.

http://hg.savannah.gnu.org/hgweb/octave/rev/56239ff815a3

Same here:

http://hg.savannah.gnu.org/hgweb/octave/rev/fafd51a1b0f0

I don't get it.  How is Mercurial keeping a record of OctJavaQry.class? 
 Does Mercurial not like the idea of merging a repository?  Bug?

Dan