miniblog.

The LLVM team have released a great document describing how to generate fast LLVM IR from your frontend: http://t.co/XvvdZeTN98

Developing Windows-friendly Python modules written in C is *hard*. http://t.co/HjjLKa2bDa Microsoft ships a compiler just for this!

Fascinating analysis of why it's hard to build a fast JIT for PHP: https://news.ycombinator.com/item?id=9123254

My Julia microbenchmark takes 20% less time to run than before (0.3.6 vs 0.3.1), and there's still scope for Julia to get faster! Exciting!

I'm a big fan of the node packaging toolchain, but I was amazed to learn they had 1 billion downloads last month! http://t.co/aA3UPdxaLE

I don't know if HTML+JS is the best way to design an editor, but I love that GitHub is exploring the boundaries of what is possible.

It's oddly satisfying to have a page where HTML rendering the bottleneck, not the backend. (The solution is lazy content loading I think.)

LLVM 3.6 is out! http://t.co/Tvi5AAeSPZ Includes a huge number of improvements.

Emacs hook tip: try to just add functions to hooks, e.g. (add-hook 'prog-mode #'eldoc-mode). It's easier to remove functions than lambdas.

Designing Qt-Style C++ APIs http://t.co/998njQunmB Full of great advice and broadly applicable.

Interesting to learn not all variables are hoisted in JS. Exceptions in catch clauses and the names of named function expressions are not.

Benign data races are scary, since C/C++ gives a compiler much freedom:

I really like the the subtitle on the npm login page. https://www.npmjs.com/login Flattery will get you everywhere.

Optimising let in JS: http://t.co/uh2zrYgk5E It's a shame that users need to consider which compilation tiers support their fave features.

"Experience with GCC is that the size of the IR is one of the most significant contributors to compile time." Wow! Not what I expected.