Correction: calling this.has(key) would only work if the has() method used hasOwnProperty().

I noticed that a lot more discussion about this has taken place in #2099, and that that issue was closed. I know that IE8 isn't exactly at the top of your list, but this is a pretty simple fix. I'd be happy to even refactor this a bit so that IE8 (using feature detection, of course) gets a different/slower version of the functions, so that this compatibility fix doesn't affect performance in modern browsers.

+1 for this issue. Would also love a simple fix to avoid breaking at load time, I confirm this works on IE8:

@@ -8146,6 +8146,7 @@
     monotone: d3_svg_lineMonotone
   });
   d3_svg_lineInterpolators.forEach(function(key, value) {
+    if (!value) return;
     value.key = key;
     value.closed = /-closed$/.test(key);
   });

FWIW, @sylvinus, aight now ships with an aight CLI tool that you can use to prevent the errant for..in iterations over __proto__ in any JS file. The included d3.ie8.js is v3.5.3, but you should be able to create an up-to-date build with:

npm install -g aight
curl -s http://d3js.org/d3.v3.js | aight >  d3.v3.ie8.js

Not that it matters, but does IE8 log __proto__ with Object.create(null) rather than Object.create({})?

For 4.0, I am currently planning on ditching the Object.create(null) for d3.map, and instead go back to the simpler prefixed properties; see src/map.js in d3-arrays. In addition to the __proto__ issue (which also existed on Node 0.10), non-prefixed keys that happened to look like numbers sometimes caused maps to behave like arrays, and then triggered large memory allocations. So forcing the keys to always look like strings rather than numbers seems like a safe choice. And the code is simpler.

So with apologies to Shawn (who I have neglected for months by not replying to this pull request), but I’m trying to minimize the amount of changes to 3.x and focus most of my effort getting 4.0 released.

No worries, Mike! IE8 support is basically off my radar now that global and government site usage is <2%. I'm also going to use custom 4.0 bundles from now on to save precious kb's, so 👍

👍