[dmd-beta] Preparing for dmd 2.066 beta

[Steven Schveighoffer via dmd-beta]

We cannot release without either reverting the changes to AA requirements, or making it so we are sure code that was flawed, but previously worked, is flagged as an error.

See this comment: https://github.com/D-Programming-Language/dmd/pull/3599#issuecomment-45275443

Code that previously (incorrectly) relies on the fact that AA code only calls compare, and defines only opCmp and not opEquals, will SILENTLY BREAK if we release today.

This is not acceptable. It needs to break, but break loudly, and not compile.

-Steve

On Jun 26, 2014, at 7:45 PM, H. S. Teoh via dmd-beta <dmd-beta at puremagic.com> wrote:

> On Thu, Jun 26, 2014 at 03:29:03PM -0700, Walter Bright via dmd-beta wrote:
>> It's time we prepared another release. Currently, we've got 11 regressions
>> to be addressed:
>> 
>> https://issues.dlang.org/buglist.cgi?bug_severity=regression&bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&list_id=47252&query_format=advanced
>> 
>> Not bad, but let's get 'em done.
> [...]
> 
> Regarding https://issues.dlang.org/show_bug.cgi?id=12255 :
> 
> I'm not sure why PR 3054 was merged, but it has some serious issues.
> First, it didn't make any sense to me to require opCmp, because AA's in
> D are unordered, so all that matters for something to be used as an AA
> key should be (1) it has a hash value, (2) two keys can be compared for
> *equality*. Requiring opCmp for AA keys just seem totally backwards to
> me. It imposes the requirement for keys to be (linearly!) orderable, but
> some AA keys need not be.
> 
> Now, I realize that the reason probably has something to do with the
> fact that the AA implementation used to call Typeinfo.compare() to
> compare AA keys, instead of Typeinfo.equals() as it ought to. But that
> has since been fixed. So the compiler really should be changed to
> require opEquals instead of opCmp, just as Steven said in the comments
> to issue 12255.
> 
> Second, PR 3054 was broken in other ways, namely issue 12905 (which
> wasn't marked as a regression, but IMO it should be).
> 
> Since we have to deal with this for the next release, let's pretty
> please do it right this time:
> 
> (1) AA keys should either have neither opEquals nor toHash, or they
> should have both (the onus is on the user to make sure they are
> consistent with each other).
> 
> (2) We should review the compiler-generated default opEquals / toHash to
> make sure they make sense, and are consistent with each other. (I
> remember seeing code in object.d where some TypeInfo's have broken
> compare/equals implementations.)
> 
> (3) Issue 12905 needs to be fixed before this release (IOW the broken
> parts of PR 3054 should be reverted).
> 
> 
> T
> 
> -- 
> What do you mean the Internet isn't filled with subliminal messages? What about all those buttons marked "submit"??
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta