Greg Hudson <ghudson@MIT.EDU> writes:
> > Unless I'm missing something basic here, it is simply not the case
> > that all directories with changed files are tied up until almost the
> > end of the edit. A dir D is tied up just exactly as long as it
> > needs to be, which is until the last postfix txdelta for a file in D
> > is handled.
>
> But postfix deltas all come at the end of the edit.
Uh, yes it's "almost the end of the edit" if one goes by number of
calls, but if one is measuring time or the amount of data flowing
through the editor, then the first postfix txdelta comes nowhere near
the end of the edit. :-)
Everything prior to the postfix txdeltas is lightweight metadata,
after all.
> >> It's not hard, it's just more complicated and error-prone, and will
> >> probably make people's lives harder in the future when it comes
> >> time to debug something related to this.
>
> > Hmm. It's hard to argue against this, as I'm sure you're aware :-).
> > I'm essentially claiming the same thing, but in reverse.
>
> Matter of opinion, I guess. I think having a couple of different
> entry points to the final cleanup and extra bookkeeping in editors
> which don't need to ref-count directories is likely to confuse
> debuggers if that code is involved in any bugs.
Mmrph. I think the special case of closing the directory whose parent
is null is pretty clear in the code, but again, I don't have a really
convincing argument for or against what you're saying.
May I ask then that we leave the change as is? (Heh, just call me
Justice Scalia.) (Sorry folks, inside joke for U.S. citizens.) If
people working with editors start to say they're confused by the
termination conditions, restoring close_edit() to the interface should
be noticeably less difficult than the original Change #1 was, and even
it wasn't so bad. But I don't see any point doing it now when things
are working and there's no clear consensus that close_edit() should
come back.
Of course, it would help if someone besides us had something to say,
but this apparently isn't a "What color should the bike shed be?" kind
of issue. :-)
-K
Received on Sat Oct 21 14:36:20 2006