More clever length estimating for import statements

Comma separated imports can get so long, they line breaks. Previously,
line breaks were inserted, if the identifier after "," made the line
longer than the soft limit. Now the whole expression length after ","
is calculated for the decision. For example, no line breaks within
"std.stdio" anymore.

src/dfmt.d

@@ -197,6 +197,28 @@ private:
                     newline();
                     break;
                 }
+                else if (current.type == tok!",")
+                {
+                    // compute length until next , or ;
+                    int length_of_next_chunk = INVALID_TOKEN_LENGTH;
+                    for (size_t i=index+1; i<tokens.length; i++)
+                    {
+                        if (tokens[i].type == tok!"," || tokens[i].type == tok!";")
+                            break;
+                        const len = tokenLength(i);
+                        assert (len >= 0);
+                        length_of_next_chunk += len;
+                    }
+                    assert (length_of_next_chunk > 0);
+                    writeToken();
+                    if (currentLineLength+1+length_of_next_chunk >= config.columnSoftLimit)
+                    {
+                        pushIndent();
+                        newline();
+                    }
+                    else
+                        write(" ");
+                }
                 else
                     formatStep();
             }