Convert negative precision to zero in printf (#1127)

and remove redundant check in grisu2_prettify.

include/fmt/format.h

@@ -1181,8 +1181,7 @@ It grisu2_prettify(const char* digits, int size, int exp, It it,
     *it++ = static_cast<Char>(params.upper ? 'E' : 'e');
     return write_exponent<Char>(exp, it);
   }
-  const int exp_threshold = 21;
-  if (size <= full_exp && full_exp <= exp_threshold) {
+  if (size <= full_exp) {
     // 1234e7 -> 12340000000[.0+]
     it = copy_str<Char>(digits, digits + size, it);
     it = std::fill_n(it, full_exp - size, static_cast<Char>('0'));

include/fmt/printf.h

@@ -44,7 +44,7 @@ class printf_precision_handler : public function<int> {
   int operator()(T value) {
     if (!int_checker<std::numeric_limits<T>::is_signed>::fits_in_int(value))
       FMT_THROW(format_error("number is too big"));
-    return static_cast<int>(value);
+    return (std::max)(static_cast<int>(value), 0);
   }
 
   template <typename T, FMT_ENABLE_IF(!std::is_integral<T>::value)>

test/printf-test.cc

@@ -520,6 +520,10 @@ TEST(PrintfTest, CheckFormatStringRegression) {
   check_format_string_regression("%c%s", 'x', "");
 }
 
+TEST(PrintfTest, FixedLargeExponent) {
+  EXPECT_EQ("1000000000000000000000", fmt::sprintf("%.*f", -13, 1e21));
+}
+
 TEST(PrintfTest, VSPrintfMakeArgsExample) {
   fmt::format_arg_store<fmt::printf_context, int, const char*> as{42,
                                                                   "something"};
@@ -599,7 +603,7 @@ std::string custom_format(const char* format_str, const Args&... args) {
   return custom_vformat(format_str, va);
 }
 
-TEST(CustomFormatterTest, Format) {
+TEST(PrintfTest, CustomFormat) {
   EXPECT_EQ("0.00", custom_format("%.2f", -.00001));
   EXPECT_EQ("0.00", custom_format("%.2f", .00001));
   EXPECT_EQ("1.00", custom_format("%.2f", 1.00001));