Eric Kow <[email protected]> writes:

> On Thu, Mar 25, 2010 at 15:58:44 +0100, Petr Rockai wrote:
>> But that's not true -- this is a problem that is not solved by
>> ignore-this. You still have to guess an encoding to use for patches that
>> don't have the ignore-this utf8 tag.  There are always to be cases where
>> this fails, no matter how you choose the encoding, so it doesn't make
>> sense to talk about correctness. We can only think of more or less
>> successful heuristics. (In your case, if the user has an utf8 locale,
>> they still get é in that patch name.)
>
> Do we all agree on the following two points?
>
> - We cannot guarantee correctness in case of untagged patches.
> - But we can guarantee it on tagged patches.
Yes, but you *can't tag preexisting patches*.

> I think you (Petr) are saying that this guarantee is actually not very
> important in practice.
I am saying that you don't get any stronger guarantee by tagging, only
that you can treat untagged things as non-utf8 by default (and *that* is
what I argue is not important in practice).

If you always try utf8 first, it is as good (i.e. guarantees
correctness) as tagging, for *new* patches, and is a very good (although
not perfect) guess for *old* patches (with a fallback on decoding
failure).

> Somewhere for the next couple of years (old software dies out slowly) or
> so, there will be an overlap period where we would not know if patches
> were created with new Darcs or with old Darcs.  During this transition
> period, would it not be desirable to at least guarantee correctness for
> the patches that are created with a new Darcs?
See above. If you treat everything as utf8, there's no difference,
tagged or untagged.

> So I can understand that that an argument based on rarity would apply
> to questions of effort-allocation and prioritisation (eg. we don't
> prioritise automated nightly builds because we think the intersection of
> people who want bleeding edge Darcs but are unwilling to build it is too
> small).  But I think it's better to avoid applying it when it comes to
> questions of trying to get the right behaviour.
Once again, this is not about right behaviour.

> I agree that the Ignore-this tagging is ugly :-(.  I'm think we're in
> arguing-about-tradeoffs territory here.  If it's not too early for me to
> in my vote, I'll fly the conservatism flag and say that we should pay
> this price of widespread ugliness, to gain that marginal improvement in
> correctness.  I say this (i) because we seem to be in one of these cases
> where when Darcs does the wrong thing, you wouldn't really know about it
> because it kinda looks right and (ii) because while we have reasons to
> make educated guesses about our users [Petr's guess is a very good one],
> I don't think we actually know that much about them.
There's no improvement in correctness. Moreover, if we want to make any
*use* of the tags, the price is not only in ugliness in metadata, but
also in (ugliness of) code and in a performance hit. Think twice...

> But if I could back off a little bit from my own conservatism, I'd say
> that what is at stake here is "only" the patch metadata (not the patches
> themselves), so it probably does not matter that much.

Yours,
   Petr.
_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to