Our nsFormSubmission and subclasses accepted a third filename argument to
explicitly specify the filename. Since switching from nsIDOMBlob to File in Bug
1085283, we can read out the filename directly from the File. This simplifies
the code, but introduces a change in the way Firefox submits form data to
servers.
Consider the code:
var fd = new FormData();
fd.append("blob1", new Blob(["hi"]), ""); // explicit empty filename as third arg
fd.append("file1", new File(["hi"], "")); // File's name is empty, no third arg.
xhr.send(fd);
Previously, the request body had filename="" in the first case, and filename="blob" in the second.
This patch will change it to both cases result in filename=""
This behaviour isn't exactly specced anywhere, nor in the HTML spec [1], nor in
RFC 2388. In addition Blink (at least Chromium v40) has the same behaviour
introduced by this patch. So shipping it seems ok to me.
[1]: http://www.w3.org/html/wg/drafts/html/master/semantics.html#multipart/form-data-encoding-algorithm
This ensures that the plaintext serializer doesn't use the preformatted
text code path if we have encountered a pre-wrap element that
Thunderbird uses (which means setting white-space: pre-wrap and width:
NNch on the body element.)
It also ensures that we use 0 as the wrap column number passed down to
the plaintext serializer, instead of -1, which this code seems to be
unable to handle properly.
This patch handles most of the call sites for these allocations except
for a few where I added TODO comments with some information. Handling
those places may require reworking lots of code, so I prefer to not do
that here.
