(Note: I've just updated my copyright assignment information, but 
haven't received confirmation that everything is in order, so this might 
need to wait until that's done for it to merge.)

There are a few related issues with pairing double quotes in CC mode 
while using `electric-pair-mode'. Hopefully the steps to reproduce below 
will explain the issues. In all the cases, I'd expect 
`electric-pair-mode' to insert a closing quote, but it doesn't. You can 
try similar steps in a `ruby-mode' buffer to see how it should work.

----------------------------------------

Common setup
------------

   $ cat foo.c
   "foobar"

   $ emacs -Q foo.c
   M-x electric-pair-mode

Note that | represents the point below.

1. Quote pairing in comments
----------------------------

   C-o   ;; to make a blank line
   // "  ;; type this

Expected: line 1 is // "|"
Actual:   line 1 is // "|

2. Inserting quote pairs before existing string
-----------------------------------------------

   "  ;; type this (point is at beginning of buffer, before "foobar")

Expected: line 1 is "|""foobar"
Actual:   line 1 is "|"foobar"

3. Splitting strings into two
-----------------------------

   "foo|bar"  ;; move point here
   "          ;; type this

Expected: line 1 is "foo"|"bar"
Actual:   line 1 is "foo"|bar"

----------------------------------------

This is because the logic in the patch for bug#36474 isn't quite right. 
Currently, `c-electric-pair-inhibit-predicate' checks if the 
newly-inserted quotation mark has "a string fence syntax-table text 
property" (i.e. if it's the start of a string literal not terminated on 
that line[1]). However, this fails in all three cases above: in (1) 
because we're in a comment, not a string literal; and in (2) and (3) 
because it's the *last* quotation mark on the line that's unterminated, 
not the one before point.

The attached patch fixes this by taking those cases into account. I also 
added `c-mode' to the list of modes to check in 
`test/lisp/electric-tests.el'. This required setting single-line 
comments as the default in those tests, since the tests expect 
single-line comments (I tried testing under `c++-mode', but the tests 
failed, I think due to <> being paren-like in C++).

[1] I think this is what it means, at least (or close to it).