Is crecord in core a DOA idea?

[Jordi Gutiérrez Hermoso]

I've started doing the work to integrate crecord into hg. It was
pretty easy to get to this stage:

    $ hg help record
    hg record [OPTION]... [FILE]...

    interactively select changes to commit

        [snip]

        If ui.interactivemode is "curses", then the interactive curses
        terminal interface will pop up. Type "?" to learn how to use
        it. See see "hg help config" for how to enable this interface.

        Otherwise, you will be prompted for whether to record changes
        to each modified file, 

        [snip]

and of course this now works. It's a simple config option that simply
changes which interface you use.

Now, after getting this simple part working, the rest of the work is a
pretty big 2-way merge (i.e. without ancestor, sadface) on the record
and crecord code. They are very similar, but there are also
significant differences. The work now consists on reconciling the two
codebases.

But, before I do that work, is there any hope for this project? The
curses interface is kind of a dead-end on Windows. Unless we make
Windows distributors ship some sort of moribund library like UniCurses
or PDCurses *in addition to* a terminal that can actually use it, the
curses interface will not work on Windows.

Can Mercurial have a Unix-only feature? Or will crecord be forever an
external extension?