jnthn just published a new blog post last night describing the fund that was recently set up for Perl 6 development (thank you, WenZPerl!) that’ll be used among other things for funding more jnthn time until the release coming in december. There’s also a lot of information on what jnthn has recently been doing in that blog post:
- Unicode database in MoarVM updated to version 7.0
- unsigned types in NativeCall (already mentioned in the previous p6weekly)
- a whole lot of RT tickets fixed, like improved error messages for BEGIN-time things and constants, “where” constraints on attributes and variables
- the beginning of the Uni class and its features, like different normalization forms (as specified by Unicode)
Nicholas Clark has been churning out patches to reduce the number of cpu instructions “used up” during rakudo’s startup and also improved the efficiency of our serialization format.
TimToady improved the MATCH method in both rakudo and nqp that are responsible for creating the high-level object that gets created for a regexp match. That’s quite a hot method and initial, noisy measurements showed it shaves off about 5% of the time spent in the “stage parse” of rakudo’s core setting.
In addition to MATCH being faster now, rakudo (and nqp) used to flip ginormous amounts of string data whenever it encountered a <?after …>, but now [correction:] if the length of the pattern inside is known, it turns into a <?before …> with the pattern itself flipped.
I’ve built support for a whole bunch of reprops (push/pop, elems, access to attributes, keys or positions, …) into the JIT for the cases where we couldn’t generate a super simple piece of assembly code to handle something (like pushing into a list that may have to enlarge the list) but we still know what exact function we’ll be calling. Unfortunately my desktop is currently not fully assembled, otherwise i’d have some benchmark data for you.
labster gave Str its own “chars” method (as opposed to inheriting it from Cool), which makes .chars on a String work about 40% faster. Wow!
Something that may have surprised a lot of people (myself included) was that Perl 6 claims that the truth value of “0” ought to be False. That is no longer the case, and so the rule for the truth value of strings is now just “does it have any characters in it?”. I think that’s more sensible.
TimToady improved the parser’s handling of heredocs in constant expressions and BEGIN blocks. There’s a bit of cheating necessary in the parser to make all of this work, but masak wrote a post on his blog about this very topic that I recommend you read.
TimToady also implemented Complex literals inside < >, which means that <3+2i>.WHAT now properly results in a Complex, rather than a Str.
Another thing TimToady did was make the “…” operator (for smart sequences) behave more correctly with regards to strings.
Thanks to lizmat, the say and note methods and subs now properly respect the “nl” setting on the IO object used.
Another thing I ought to mention is the stellar ticket and testing work that bartolin has been doing for what feels like many months.
There’s now a web site examples.perl6.org with lots of example code for you to peruse!
One other thing worth mentioning is that you now get a perl6-gdb-m and perl6-valgrind-m script installed with your rakudo for quickly running your stuff under gdb or valgrind.
And with that I’m already finished with the weekly. This time it’s done before midnight local time, too!
I wish you a pleasant week; until next time!