Incomplete XHR responses from IIS with compression

Question

I'm loading scripts with jQuery ajax.

Here's a snip from the loader code, for what it's worth.

var resources = ['knockout-3.2.0.js', 'fonts/fontawesome-webfont.eot', ... ];
var retryCount = 0;
load(0);
function load(i) {
  if (i == resources.length) {
    $("#progBar").width("100%");
    $("#progCaption").text("executing App/main...")
    var requireScript = document.createElement('script');
    requireScript.setAttribute("type", "text/javascript");
    requireScript.setAttribute("src", "Scripts/require.js");
    requireScript.setAttribute("data-main", "App/main");
    head[0].appendChild(requireScript);
  }
  else {
    $("#progBar").width(100 * i / resources.length + "%");
    var r = resources[i];
    switch (r.substring(r.lastIndexOf('.'))) {
      case ".js":
        $("#progCaption").text(r);
        $.getScript('Scripts/' + r)
        .done(function (data) {
          retryCount = 0;
          load(++i);
        })
        .error(function (err) {
          if (err.status == 404 || retryCount == 3) {
            load(++i);
          } else {
            retryCount++;
            console.log('Retry ' + r);
            load(i);
          }
        });
        break;
      ...

Recommended syntax for getScript only takes one parameter. Either it's a valid URL or not. Because the start of the resource is returned we know the URL is valid . There is no error in the invoking code, jQuery is failing to get the entire response before parsing it as a script.

Pointed at the same public facing server, $.getScript() throws an error even though err.status reports 200 OK. Inspection of the data shows that it is truncated, and the error is probably a JS parse failure as a result of the truncation.

This only happens with IE11 at work, and the problem stops manifesting when the traffic is routed through Fiddler. It doesn't occur at all on IE11 at home. Changing 'knockout-3.2.0.js', to 'knockout-3.2.0.debug.js', causes knockout to load without a hitch, however bootstrap.js isn't minified and it does play up, so it's not as simple as no line breaks. I thought it might be some kind of server caching issue fixed by the name change (it can't be a browser caching problem since getScript applies a cache-buster parameter) but I tried renaming knockout-3.2.0.js to knockout-3.2.0.min.js and it made no difference.

Questions with similar symptoms:

• IE returns incomplete HTML via Ajax This is similar but IE11 is hardly an "older" version, and adding Fiddler to the equation fixes it so the problem is unlikely to be a parsing hiccup.

• Ajax request returns 200 OK, but an error event is fired instead of success In this case the parsing failure is caused by inappropriate typing of the response, not an incomplete response.

---

I verified that IE at home and at work were both 11.0.9600.17690 update KB3032359.

The problem vanished when I disabled IIS compression.

I don't know whether it was static or dynamic compression because I turned them off at the same time.

If anyone from Microsoft wishes to investigate (I wouldn't be surprised if the answer is no, it's a bit of an edge case)... I have PCAP files of the failing exchange between client and server.

Diagnostic summary

• it's not the network, it fails with an alternate route through my phone
• it's not the version of IE, same IE build at home does not fail
• it's not the platform (both on Win81) although it could be network hardware drivers
• it's not IIS, works at home and on my WinPhone81 and for all other browsers
• it is IIS, turning off server compression fixes it
• it's not IIS, compression worked fine for other browsers, although I didn't think to capture the traffic with a browser that worked with compression on
• interposing Fiddler resolved the problem while compression was active
• it's not a browser caching problem, getScript supplies a cache-buster parameter and when I deliberately allowed caching it was obvious in the capture files as http no-change responses.

I have the oddest feeling this is related Why does IE11 create blank post request except when Fiddler is running? but I couldn't say why.

Once the files are in cache their size ceases to signify, which why I run a pre-loader - for a SPA like mine it's effectively installation. A powerful argument for SPA over traditional application architectures. But lengthy setup does somewhat mar the first time experience.

Answer 1

The problem is resolved by disabling compression in IIS.