Interned pathnames propagating bad information

[MOON at SCRC-TENEX]

    Date: Monday, 22 March 1982, 04:10-EST
    From: Robert W. Kerns <RWK at SCRC-TENEX>
    Subject: Interned pathnames propagating bad information
    To: MOON at SCRC-TENEX
    Cc: BUG-LISPM at MIT-AI, BUG-ZWEI at SCRC-TENEX, turnip at MIT-AI
    In-reply-to: The message of 21 Mar 82 23:40-EST from MOON at SCRC-TENEX

        Date: Sunday, 21 March 1982  23:40-EST
        From: MOON at SCRC-TENEX
    	....
        If it changed the -*- line I would have been shafted about 3 times a day
        by it during the past several weeks.  I don't think it should ever change
        anything about the file either; you only want that behavior because of
        the bug now that Zwei's assumption of a package when none is specified
        gets speciously associated with the file and then you need to undo that
        specious package property.
    Can you explain how you would have been shafted?  I don't see it.  Maybe
    I have a different model of what Set Package means or something.

    I agree it doesn't matter as much if ZWEI didn't remember properties it
    spuriously assumed.

I have been working with a lot of files that temporarily have to exist in
more than one package.  m-x Set Package is the only way to set what package
c-sh-E and c-sh-C use when reading from the buffer.  If it automatically
updated the -*- line I would be totally and completely shafted, because the
buffer would be modified spuriously, and because the file's real package
would be suddenly replaced by the temporary other package.  Having it merely
update the in-core copy of the file's -*- line is mostly not a shaft, since
the -*- line will set it back the next time the file is compiled or loaded
(except in the screw case that was the source of the original bug report,
where someone omitted the Package: in the file's -*- line on the theory that
it would always default to User.)

To get to the point, there are good reasons to have two copies of this
information, and I think the copies should be more separate than they are
now.  Possibly a better user interface would be to have two Set Package
commands, one which updated both copies and one which updated only one
copy, instead of one (half-assed) command that updates part but not all
of both copies.  But the assumption that any particular will only be read
into one package is highly naive.