2015.12: A release. Also, lazy deserialization makes a comeback

On Friday, the 20th we got a Rakudo release, followed swiftly by a Star release on the 21st and a Rakudo Star .msi installer this monday. There’s also one Rakudo Star Docker Image based on Fedora 21 by mj41 and one Rakudo Star Docker Image based on Ubuntu 14.04 by hoelzro. On top of that, FROGGS wrote down how to create the Rakudo Star .msi packages. Big thanks to FROGGS for the Rakudo release and the MSI, moritz for the Rakudo Star release and everyone else involved in the Release process and all the tasks surrounding it!

Performance

A long time ago, MoarVM learned how to delay deserialization of a whole bunch of objects until they are referred to, but there was at least one bug that was caused by it and so it was disabled for a long while. That feature was re-enabled just after the release, so that we get quite some testing by the community to find out if it still causes issues and this time hopefully get rid of them, since the lazy deserialization gives us much better memory usage and speed when starting up a program that doesn’t use every class there is.

That’s not the only change we can look forward to, though. Jnthn is currently working on improving the impact of Frames on the garbage collector a whole lot, which (in my expectation) should improve memory usage as well as collection times.

In addition, Jnthn will fix my attempts at teaching NQP’s code-gen and Rakudo’s optimizer about references to local variables, which will recover the performance hit we took when our support for native variables and attributes has been vastly improved recently.

There was also a commit that prevented $*VM from being initialized on every startup that dramatically decreased startup time, but that caused a problem with installing modules, so it was reverted. Maybe we’ll end up getting it back this month with a proper fix for the underlying problem!

There’s also two things yours truly is working on: Making calls to a repr’s operations (attribute access, pushing/popping, positional and associative accesses, …) less overhead-ridden in the JIT compiler and merging facts about registers in spesh when Blocks in the Command Flow Graph come together. They aren’t ready for merging yet and also don’t seem to cause significant performance improvements, but I’ll keep looking for ways to make things faster …

Platform things

Rakudo used to follow parrot’s install directory layout, which put things in $PREFIX/languages. We changed that to $PREFIX/share instead; you may have to re-install modules and run Configure.pl again for all parts of your Rakudo installation if you’re staying on the bleeding edge via a git checkout. If things get very wonky with rakudobrew, consider nuking the subdirectories of ~/.rakudobrew …

This change was made to make rakudo and nqp much more friendly towards linux distributions.

Also, the docker image efforts by hoelzro and mj41 will be combined into a more official rakudo star docker repository.

What’s also very worth mentioning is Coke’s ongoing work to get our stuff into the official MacPorts repository. There’s only MoarVM so far, but NQP and Rakudo shall follow.

Language Features, Syntax, Bug fixes

usev6 fixed the interaction of “last” and the sequence operator, making sure that code like this

perl6 -e 'say 10, 9, 8, { $_ - 1 || last } ... *'
10 9 8 7 6 5 4 3 2 1 1

now properly drops the last entry in the list.

Raydiak (and Mouq) implemented the simplest case of an angle-bracketed Rat literal, like <1/5>. In the future, there’ll also be support for lists of Rat literals like <1/2 1/3 1/4>. I may have to bug them to also make Complex literals work in angle brackets 🙂

Psch fixed a problem in Rakudo-JVM’s Binder that’d cause coercive types to pass the original value rather than the coerced value to a function’s body. After that, he also implemented nqp::getuniname on Rakudo-JVM which powers Perl6’s uniname sub. A few improvements for the Eval Server have also landed.

Thanks to TimToady the “cannot call $sub_or_method, none of these signatures match:” error message also shows the capture that was passed and caused the “no match” exception. The robustness of Mu’s .perl method has also been improved and its output prettified a tiny bit.

Lastly, Mouq fixed the combination of Bag and hyper operators, such as “Bag.new(…) »-» 1”.

Another thing happening in a non-merged branch right now is more features and improvements for supporting CUnion meant for (but not limited to) NativeCall by FROGGS.

Documentation

moritz, ptc and others have been improving the official doc.perl6.org contents again. That’s Way Cool if you ask me! As always: feedback is very welcome on the IRC channel or the perl6-users mailing list, or even in the comments section of this blog.

Finishing up

I feel like I’ve missed a few things, but I’ve been procrastinating this post far too long now and bedtime is coming up soon 🙂

Therefor I wish you a nice week and have fun trying out the newest Rakudo Star release!

P.S.: I’m thinking about adding a second series of posts to this blog. Once a week I’d look through all the snippets evaluated with our evalbot on the IRC channel and pick out one or two and briefly describe what I like about them. What do you think about that idea?

Advertisements

4 thoughts on “2015.12: A release. Also, lazy deserialization makes a comeback

  1. Hi Timotimo!
    Snippet reviews would be great! I look forward to see interesting new stuff.

    Greetings,
    Perl6_newbee

Got something to note?

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s