2014-07-25 05:31:46 +08:00
|
|
|
#include <algorithm>
|
2015-10-17 06:35:11 +08:00
|
|
|
#include <regex>
|
2014-07-25 05:31:46 +08:00
|
|
|
|
2015-11-21 09:41:32 +08:00
|
|
|
#include <xlnt/utils/datetime.hpp>
|
2015-11-03 21:38:09 +08:00
|
|
|
#include <xlnt/utils/hash_combine.hpp>
|
2014-08-14 06:56:34 +08:00
|
|
|
#include <xlnt/styles/number_format.hpp>
|
2014-07-24 04:00:09 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
namespace {
|
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
const std::unordered_map<std::size_t, std::string> &builtin_formats()
|
2015-11-03 05:45:05 +08:00
|
|
|
{
|
2015-11-11 07:58:54 +08:00
|
|
|
static const std::unordered_map<std::size_t, std::string> *formats =
|
|
|
|
new std::unordered_map<std::size_t, std::string>
|
2015-11-03 05:45:05 +08:00
|
|
|
({
|
|
|
|
{ 0, "General" },
|
|
|
|
{ 1, "0" },
|
|
|
|
{ 2, "0.00" },
|
|
|
|
{ 3, "#,##0" },
|
|
|
|
{ 4, "#,##0.00" },
|
|
|
|
{ 5, "\"$\"#,##0_);(\"$\"#,##0)" },
|
|
|
|
{ 6, "\"$\"#,##0_);[Red](\"$\"#,##0)" },
|
|
|
|
{ 7, "\"$\"#,##0.00_);(\"$\"#,##0.00)" },
|
|
|
|
{ 8, "\"$\"#,##0.00_);[Red](\"$\"#,##0.00)" },
|
|
|
|
{ 9, "0%" },
|
|
|
|
{ 10, "0.00%" },
|
|
|
|
{ 11, "0.00E+00" },
|
|
|
|
{ 12, "# ?/?" },
|
|
|
|
{ 13, "# \?\?/??" }, // escape trigraph
|
|
|
|
{ 14, "mm-dd-yy" },
|
|
|
|
{ 15, "d-mmm-yy" },
|
|
|
|
{ 16, "d-mmm" },
|
|
|
|
{ 17, "mmm-yy" },
|
|
|
|
{ 18, "h:mm AM/PM" },
|
|
|
|
{ 19, "h:mm:ss AM/PM" },
|
|
|
|
{ 20, "h:mm" },
|
|
|
|
{ 21, "h:mm:ss" },
|
|
|
|
{ 22, "m/d/yy h:mm" },
|
|
|
|
|
|
|
|
{ 37, "#,##0_);(#,##0)" },
|
|
|
|
{ 38, "#,##0_);[Red](#,##0)" },
|
|
|
|
{ 39, "#,##0.00_);(#,##0.00)" },
|
|
|
|
{ 40, "#,##0.00_);[Red](#,##0.00)" },
|
|
|
|
|
|
|
|
{ 41, "_(* #,##0_);_(* \\(#,##0\\);_(* \"-\"_);_(@_)" },
|
|
|
|
{ 42, "_(\"$\"* #,##0_);_(\"$\"* \\(#,##0\\);_(\"$\"* \"-\"_);_(@_)" },
|
|
|
|
{ 43, "_(* #,##0.00_);_(* \\(#,##0.00\\);_(* \"-\"??_);_(@_)" },
|
|
|
|
|
|
|
|
{ 44, "_(\"$\"* #,##0.00_)_(\"$\"* \\(#,##0.00\\)_(\"$\"* \"-\"??_)_(@_)" },
|
|
|
|
{ 45, "mm:ss" },
|
|
|
|
{ 46, "[h]:mm:ss" },
|
|
|
|
{ 47, "mmss.0" },
|
|
|
|
{ 48, "##0.0E+0" },
|
|
|
|
{ 49, "@" }
|
|
|
|
|
|
|
|
// EXCEL differs from the standard in the following:
|
|
|
|
//{14, "m/d/yyyy"},
|
|
|
|
//{22, "m/d/yyyy h:mm"},
|
|
|
|
//{37, "#,##0_);(#,##0)"},
|
|
|
|
//{38, "#,##0_);[Red]"},
|
|
|
|
//{39, "#,##0.00_);(#,##0.00)"},
|
|
|
|
//{40, "#,##0.00_);[Red]"},
|
|
|
|
//{47, "mm:ss.0"},
|
|
|
|
//{55, "yyyy/mm/dd"}
|
|
|
|
});
|
|
|
|
|
|
|
|
return *formats;
|
2014-08-02 04:46:54 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-11-21 09:41:32 +08:00
|
|
|
enum class condition_type
|
|
|
|
{
|
|
|
|
less_than,
|
|
|
|
less_or_equal,
|
|
|
|
equal,
|
|
|
|
greater_than,
|
|
|
|
greater_or_equal,
|
|
|
|
invalid
|
|
|
|
};
|
|
|
|
|
|
|
|
struct section
|
|
|
|
{
|
|
|
|
bool has_value = false;
|
|
|
|
std::string value;
|
|
|
|
bool has_color = false;
|
|
|
|
std::string color;
|
|
|
|
bool has_condition = false;
|
|
|
|
condition_type condition = condition_type::invalid;
|
|
|
|
std::string condition_value;
|
|
|
|
bool has_locale = false;
|
|
|
|
std::string locale;
|
|
|
|
|
|
|
|
section &operator=(const section &other)
|
|
|
|
{
|
|
|
|
has_value = other.has_value;
|
|
|
|
value = other.value;
|
|
|
|
has_color = other.has_color;
|
|
|
|
color = other.color;
|
|
|
|
has_condition = other.has_condition;
|
|
|
|
condition = other.condition;
|
|
|
|
condition_value = other.condition_value;
|
|
|
|
has_locale = other.has_locale;
|
|
|
|
locale = other.locale;
|
|
|
|
|
|
|
|
return *this;
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
struct format_sections
|
|
|
|
{
|
|
|
|
section first;
|
|
|
|
section second;
|
|
|
|
section third;
|
|
|
|
section fourth;
|
|
|
|
};
|
|
|
|
|
|
|
|
// copied from named_range.cpp, keep in sync
|
|
|
|
/// <summary>
|
|
|
|
/// Return a vector containing string split at each delim.
|
|
|
|
/// </summary>
|
|
|
|
/// <remark>
|
|
|
|
/// This should maybe be in a utility header so it can be used elsewhere.
|
|
|
|
/// </remarks>
|
|
|
|
std::vector<std::string> split_string(const std::string &string, char delim)
|
|
|
|
{
|
|
|
|
std::vector<std::string> split;
|
|
|
|
std::string::size_type previous_index = 0;
|
|
|
|
auto separator_index = string.find(delim);
|
|
|
|
|
|
|
|
while (separator_index != std::string::npos)
|
|
|
|
{
|
|
|
|
auto part = string.substr(previous_index, separator_index - previous_index);
|
|
|
|
split.push_back(part);
|
|
|
|
|
|
|
|
previous_index = separator_index + 1;
|
|
|
|
separator_index = string.find(delim, previous_index);
|
|
|
|
}
|
|
|
|
|
|
|
|
split.push_back(string.substr(previous_index));
|
|
|
|
|
|
|
|
return split;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<std::string> split_string_any(const std::string &string, const std::string &delims)
|
|
|
|
{
|
|
|
|
std::vector<std::string> split;
|
|
|
|
std::string::size_type previous_index = 0;
|
|
|
|
auto separator_index = string.find_first_of(delims);
|
|
|
|
|
|
|
|
while (separator_index != std::string::npos)
|
|
|
|
{
|
|
|
|
auto part = string.substr(previous_index, separator_index - previous_index);
|
|
|
|
|
|
|
|
if (!part.empty())
|
|
|
|
{
|
|
|
|
split.push_back(part);
|
|
|
|
}
|
|
|
|
|
|
|
|
previous_index = separator_index + 1;
|
|
|
|
separator_index = string.find_first_of(delims, previous_index);
|
|
|
|
}
|
|
|
|
|
|
|
|
split.push_back(string.substr(previous_index));
|
|
|
|
|
|
|
|
return split;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool is_date_format(const std::string &format_string)
|
|
|
|
{
|
|
|
|
auto not_in = format_string.find_first_not_of("/-:, mMyYdDhHsS");
|
|
|
|
return not_in == std::string::npos;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool is_valid_color(const std::string &color)
|
|
|
|
{
|
|
|
|
static const std::vector<std::string> *colors =
|
|
|
|
new std::vector<std::string>(
|
|
|
|
{
|
|
|
|
"Black",
|
|
|
|
"Green"
|
|
|
|
"White",
|
|
|
|
"Blue",
|
|
|
|
"Magenta",
|
|
|
|
"Yellow",
|
|
|
|
"Cyan",
|
|
|
|
"Red"
|
|
|
|
});
|
|
|
|
|
|
|
|
auto compare_color = [&](const std::string &other) {
|
|
|
|
if (color.size() != other.size()) return false;
|
|
|
|
|
|
|
|
for (std::size_t i = 0; i < color.size(); i++)
|
|
|
|
{
|
|
|
|
if (std::toupper(color[i]) != std::toupper(other[i]))
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
return std::find_if(colors->begin(), colors->end(), compare_color) != colors->end();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool parse_condition(const std::string &string, section &s)
|
|
|
|
{
|
|
|
|
s.has_condition = false;
|
|
|
|
s.condition = condition_type::invalid;
|
|
|
|
s.condition_value.clear();
|
|
|
|
|
|
|
|
if (string[0] == '<')
|
|
|
|
{
|
|
|
|
s.has_condition = true;
|
|
|
|
|
|
|
|
if (string[1] == '=')
|
|
|
|
{
|
|
|
|
s.condition = condition_type::less_or_equal;
|
|
|
|
s.condition_value = string.substr(2);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
s.condition = condition_type::less_than;
|
|
|
|
s.condition_value = string.substr(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (string[0] == '>')
|
|
|
|
{
|
|
|
|
s.has_condition = true;
|
|
|
|
|
|
|
|
if (string[1] == '=')
|
|
|
|
{
|
|
|
|
s.condition = condition_type::greater_or_equal;
|
|
|
|
s.condition_value = string.substr(2);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
s.condition = condition_type::greater_than;
|
|
|
|
s.condition_value = string.substr(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (string[0] == '=')
|
|
|
|
{
|
|
|
|
s.has_condition = true;
|
|
|
|
s.condition = condition_type::equal;
|
|
|
|
s.condition_value = string.substr(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
return s.has_condition;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool is_hex(char c)
|
|
|
|
{
|
|
|
|
if (c >= 'A' || c <= 'F') return true;
|
|
|
|
if (c >= '0' || c <= '9') return true;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::unordered_map<int, std::string> known_locales()
|
|
|
|
{
|
|
|
|
const std::unordered_map<int, std::string> *all =
|
|
|
|
new std::unordered_map<int, std::string>(
|
|
|
|
{
|
|
|
|
{ 0x401, "Arabic - Saudi Arabia" },
|
|
|
|
{ 0x402, "Bulgarian" },
|
|
|
|
{ 0x403, "Catalan" },
|
|
|
|
{ 0x404, "Chinese - Taiwan" },
|
|
|
|
{ 0x405, "Czech" },
|
|
|
|
{ 0x406, "Danish" },
|
|
|
|
{ 0x407, "German - Germany" },
|
|
|
|
{ 0x408, "Greek" },
|
|
|
|
{ 0x409, "English - United States" },
|
|
|
|
{ 0x410, "Italian - Italy" },
|
|
|
|
{ 0x411, "Japanese" },
|
|
|
|
{ 0x412, "Korean" },
|
|
|
|
{ 0x413, "Dutch - Netherlands" },
|
|
|
|
{ 0x414, "Norwegian - Bokml" },
|
|
|
|
{ 0x415, "Polish" },
|
|
|
|
{ 0x416, "Portuguese - Brazil" },
|
|
|
|
{ 0x417, "Raeto-Romance" },
|
|
|
|
{ 0x418, "Romanian - Romania" },
|
|
|
|
{ 0x419, "Russian" },
|
|
|
|
{ 0x420, "Urdu" },
|
|
|
|
{ 0x421, "Indonesian" },
|
|
|
|
{ 0x422, "Ukrainian" },
|
|
|
|
{ 0x423, "Belarusian" },
|
|
|
|
{ 0x424, "Slovenian" },
|
|
|
|
{ 0x425, "Estonian" },
|
|
|
|
{ 0x426, "Latvian" },
|
|
|
|
{ 0x427, "Lithuanian" },
|
|
|
|
{ 0x428, "Tajik" },
|
|
|
|
{ 0x429, "Farsi - Persian" },
|
|
|
|
{ 0x430, "Sesotho (Sutu)" },
|
|
|
|
{ 0x431, "Tsonga" },
|
|
|
|
{ 0x432, "Setsuana" },
|
|
|
|
{ 0x433, "Venda" },
|
|
|
|
{ 0x434, "Xhosa" },
|
|
|
|
{ 0x435, "Zulu" },
|
|
|
|
{ 0x436, "Afrikaans" },
|
|
|
|
{ 0x437, "Georgian" },
|
|
|
|
{ 0x438, "Faroese" },
|
|
|
|
{ 0x439, "Hindi" },
|
|
|
|
{ 0x440, "Kyrgyz - Cyrillic" },
|
|
|
|
{ 0x441, "Swahili" },
|
|
|
|
{ 0x442, "Turkmen" },
|
|
|
|
{ 0x443, "Uzbek - Latin" },
|
|
|
|
{ 0x444, "Tatar" },
|
|
|
|
{ 0x445, "Bengali - India" },
|
|
|
|
{ 0x446, "Punjabi" },
|
|
|
|
{ 0x447, "Gujarati" },
|
|
|
|
{ 0x448, "Oriya" },
|
|
|
|
{ 0x449, "Tamil" },
|
|
|
|
{ 0x450, "Mongolian" },
|
|
|
|
{ 0x451, "Tibetan" },
|
|
|
|
{ 0x452, "Welsh" },
|
|
|
|
{ 0x453, "Khmer" },
|
|
|
|
{ 0x454, "Lao" },
|
|
|
|
{ 0x455, "Burmese" },
|
|
|
|
{ 0x456, "Galician" },
|
|
|
|
{ 0x457, "Konkani" },
|
|
|
|
{ 0x458, "Manipuri" },
|
|
|
|
{ 0x459, "Sindhi" },
|
|
|
|
{ 0x460, "Kashmiri" },
|
|
|
|
{ 0x461, "Nepali" },
|
|
|
|
{ 0x462, "Frisian - Netherlands" },
|
|
|
|
{ 0x464, "Filipino" },
|
|
|
|
{ 0x465, "Divehi; Dhivehi; Maldivian" },
|
|
|
|
{ 0x466, "Edo" },
|
|
|
|
{ 0x470, "Igbo - Nigeria" },
|
|
|
|
{ 0x474, "Guarani - Paraguay" },
|
|
|
|
{ 0x476, "Latin" },
|
|
|
|
{ 0x477, "Somali" },
|
|
|
|
{ 0x481, "Maori" },
|
|
|
|
{ 0x801, "Arabic - Iraq" },
|
|
|
|
{ 0x804, "Chinese - China" },
|
|
|
|
{ 0x807, "German - Switzerland" },
|
|
|
|
{ 0x809, "English - Great Britain" },
|
|
|
|
{ 0x810, "Italian - Switzerland" },
|
|
|
|
{ 0x813, "Dutch - Belgium" },
|
|
|
|
{ 0x814, "Norwegian - Nynorsk" },
|
|
|
|
{ 0x816, "Portuguese - Portugal" },
|
|
|
|
{ 0x818, "Romanian - Moldova" },
|
|
|
|
{ 0x819, "Russian - Moldova" },
|
|
|
|
{ 0x843, "Uzbek - Cyrillic" },
|
|
|
|
{ 0x845, "Bengali - Bangladesh" },
|
|
|
|
{ 0x850, "Mongolian" },
|
|
|
|
{ 0x1001, "Arabic - Libya" },
|
|
|
|
{ 0x1004, "Chinese - Singapore" },
|
|
|
|
{ 0x1007, "German - Luxembourg" },
|
|
|
|
{ 0x1009, "English - Canada" },
|
|
|
|
{ 0x1401, "Arabic - Algeria" },
|
|
|
|
{ 0x1404, "Chinese - Macau SAR" },
|
|
|
|
{ 0x1407, "German - Liechtenstein" },
|
|
|
|
{ 0x1409, "English - New Zealand" },
|
|
|
|
{ 0x1801, "Arabic - Morocco" },
|
|
|
|
{ 0x1809, "English - Ireland" },
|
|
|
|
{ 0x2001, "Arabic - Oman" },
|
|
|
|
{ 0x2009, "English - Jamaica" },
|
|
|
|
{ 0x2401, "Arabic - Yemen" },
|
|
|
|
{ 0x2409, "English - Caribbean" },
|
|
|
|
{ 0x2801, "Arabic - Syria" },
|
|
|
|
{ 0x2809, "English - Belize" },
|
|
|
|
{ 0x3001, "Arabic - Lebanon" },
|
|
|
|
{ 0x3009, "English - Zimbabwe" },
|
|
|
|
{ 0x3401, "Arabic - Kuwait" },
|
|
|
|
{ 0x3409, "English - Phillippines" },
|
|
|
|
{ 0x3801, "Arabic - United Arab Emirates" },
|
|
|
|
{ 0x4001, "Arabic - Qatar" }
|
|
|
|
});
|
|
|
|
|
|
|
|
return *all;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool is_valid_locale(const std::string &locale_string)
|
|
|
|
{
|
|
|
|
std::string country = locale_string.substr(locale_string.find('-') + 1);
|
|
|
|
|
|
|
|
if (country.empty())
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (auto c : country)
|
|
|
|
{
|
|
|
|
if (!is_hex(static_cast<char>(std::toupper(c))))
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
auto index = std::stoi(country, 0, 16);
|
|
|
|
|
|
|
|
auto known_locales_ = known_locales();
|
|
|
|
|
|
|
|
if (known_locales_.find(index) == known_locales_.end())
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string beginning = locale_string.substr(0, locale_string.find('-'));
|
|
|
|
|
|
|
|
if (beginning.empty() || beginning[0] != '$')
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (beginning.size() == 1)
|
|
|
|
{
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
beginning = beginning.substr(1);
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
section parse_section(const std::string §ion_string)
|
|
|
|
{
|
|
|
|
section s;
|
|
|
|
|
|
|
|
std::string format_part;
|
|
|
|
std::string bracket_part;
|
|
|
|
|
|
|
|
std::vector<std::string> bracket_parts;
|
|
|
|
|
|
|
|
bool in_quotes = false;
|
|
|
|
bool in_brackets = false;
|
|
|
|
|
|
|
|
const std::vector<std::string> bracket_times = { "h", "hh", "m", "mm", "s", "ss" };
|
|
|
|
|
|
|
|
for (std::size_t i = 0; i < section_string.size(); i++)
|
|
|
|
{
|
|
|
|
if (!in_quotes && section_string[i] == '"')
|
|
|
|
{
|
|
|
|
format_part.push_back(section_string[i]);
|
|
|
|
in_quotes = true;
|
|
|
|
}
|
|
|
|
else if (in_quotes && section_string[i] == '"')
|
|
|
|
{
|
|
|
|
format_part.push_back(section_string[i]);
|
|
|
|
|
|
|
|
if (i < section_string.size() - 1 && section_string[i + 1] != '"')
|
|
|
|
{
|
|
|
|
in_quotes = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (!in_brackets && section_string[i] == '[')
|
|
|
|
{
|
|
|
|
in_brackets = true;
|
|
|
|
|
|
|
|
for (auto bracket_time : bracket_times)
|
|
|
|
{
|
|
|
|
if (i < section_string.size() - bracket_time.size() &&
|
|
|
|
section_string.substr(i + 1, bracket_time.size()) == bracket_time)
|
|
|
|
{
|
|
|
|
in_brackets = false;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (in_brackets)
|
|
|
|
{
|
|
|
|
if (section_string[i] == ']')
|
|
|
|
{
|
|
|
|
in_brackets = false;
|
|
|
|
|
|
|
|
if (is_valid_color(bracket_part))
|
|
|
|
{
|
|
|
|
if (s.color.empty())
|
|
|
|
{
|
|
|
|
s.color = bracket_part;
|
|
|
|
s.has_color = true;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
throw std::runtime_error("two colors");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (is_valid_locale(bracket_part))
|
|
|
|
{
|
|
|
|
if (s.locale.empty())
|
|
|
|
{
|
|
|
|
s.locale = bracket_part;
|
|
|
|
s.has_locale = true;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
throw std::runtime_error("two locales");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (s.has_condition || !parse_condition(bracket_part, s))
|
|
|
|
{
|
|
|
|
throw std::runtime_error("invalid bracket format");
|
|
|
|
}
|
|
|
|
|
|
|
|
bracket_part.clear();
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
bracket_part.push_back(section_string[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
format_part.push_back(section_string[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
s.value = format_part;
|
|
|
|
s.has_value = true;
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
format_sections parse_format_sections(const std::string &combined)
|
|
|
|
{
|
|
|
|
format_sections result = {};
|
|
|
|
|
|
|
|
auto split = split_string(combined, ';');
|
|
|
|
|
|
|
|
if (split.empty())
|
|
|
|
{
|
|
|
|
throw std::runtime_error("empty string");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.first = parse_section(split[0]);
|
|
|
|
|
|
|
|
if (!result.first.has_condition)
|
|
|
|
{
|
|
|
|
result.second = result.first;
|
|
|
|
result.third = result.first;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (split.size() > 1)
|
|
|
|
{
|
|
|
|
result.second = parse_section(split[1]);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (split.size() > 2)
|
|
|
|
{
|
|
|
|
if (result.first.has_condition && !result.second.has_condition)
|
|
|
|
{
|
|
|
|
throw std::runtime_error("first two sections should have conditions");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.third = parse_section(split[2]);
|
|
|
|
|
|
|
|
if (result.third.has_condition)
|
|
|
|
{
|
|
|
|
throw std::runtime_error("third section shouldn't have a condition");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (split.size() > 3)
|
|
|
|
{
|
|
|
|
if (result.first.has_condition)
|
|
|
|
{
|
|
|
|
throw std::runtime_error("too many parts");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.fourth = parse_section(split[3]);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (split.size() > 4)
|
|
|
|
{
|
|
|
|
throw std::runtime_error("too many parts");
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string format_section(long double number, const section &format, xlnt::calendar base_date)
|
|
|
|
{
|
|
|
|
const std::string unquoted = "$+(:^'{<=-/)!&~}> ";
|
|
|
|
std::string format_temp = format.value;
|
|
|
|
std::string result;
|
|
|
|
|
|
|
|
if (is_date_format(format.value))
|
|
|
|
{
|
|
|
|
const std::string date_unquoted = ",-/: ";
|
|
|
|
|
|
|
|
const std::vector<std::string> dates = { "m", "mm", "mmm", "mmmmm", "mmmmmm", "d", "dd", "ddd", "dddd", "yy",
|
|
|
|
"yyyy", "h", "[h]", "hh", "m", "[m]", "mm", "s", "[s]", "ss", "AM/PM",
|
|
|
|
"am/pm", "A/P", "a/p" };
|
|
|
|
|
|
|
|
const std::vector<std::string> MonthNames = { "January", "February", "March",
|
|
|
|
"April", "May", "June", "July", "August", "September", "October", "November", "December" };
|
|
|
|
|
|
|
|
|
|
|
|
auto split = split_string_any(format.value, date_unquoted);
|
|
|
|
std::string::size_type index = 0, prev = 0;
|
|
|
|
auto d = xlnt::datetime::from_number(number, base_date);
|
|
|
|
bool processed_month = false;
|
|
|
|
|
|
|
|
auto lower_string = [](const std::string &s) {
|
|
|
|
std::string lower;
|
|
|
|
lower.resize(s.size());
|
|
|
|
for (std::size_t i = 0; i < s.size(); i++)
|
|
|
|
lower[i] = static_cast<char>(std::tolower(s[i]));
|
|
|
|
return lower;
|
|
|
|
};
|
|
|
|
|
|
|
|
for (auto part : split)
|
|
|
|
{
|
|
|
|
while (format.value.substr(index, part.size()) != part)
|
|
|
|
{
|
|
|
|
index++;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto between = format.value.substr(prev, index - prev);
|
|
|
|
result.append(between);
|
|
|
|
|
|
|
|
part = lower_string(part);
|
|
|
|
|
|
|
|
if (part == "m" && !processed_month)
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.month));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "mm" && !processed_month)
|
|
|
|
{
|
|
|
|
if (d.month < 10)
|
|
|
|
{
|
|
|
|
result.append("0");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.append(std::to_string(d.month));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "mmm" && !processed_month)
|
|
|
|
{
|
|
|
|
result.append(MonthNames.at(static_cast<std::size_t>(d.month - 1)).substr(0, 3));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "mmmm" && !processed_month)
|
|
|
|
{
|
|
|
|
result.append(MonthNames.at(static_cast<std::size_t>(d.month - 1)));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "d")
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.day));
|
|
|
|
}
|
|
|
|
else if (part == "dd")
|
|
|
|
{
|
|
|
|
if (d.day < 10)
|
|
|
|
{
|
|
|
|
result.append("0");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.append(std::to_string(d.day));
|
|
|
|
}
|
|
|
|
else if (part == "yyyy")
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.year));
|
|
|
|
}
|
|
|
|
|
|
|
|
else if (part == "h")
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.hour));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "hh")
|
|
|
|
{
|
|
|
|
if (d.hour < 10)
|
|
|
|
{
|
|
|
|
result.append("0");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.append(std::to_string(d.hour));
|
|
|
|
processed_month = true;
|
|
|
|
}
|
|
|
|
else if (part == "m")
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.minute));
|
|
|
|
}
|
|
|
|
else if (part == "mm")
|
|
|
|
{
|
|
|
|
if (d.minute < 10)
|
|
|
|
{
|
|
|
|
result.append("0");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.append(std::to_string(d.minute));
|
|
|
|
}
|
|
|
|
else if (part == "s")
|
|
|
|
{
|
|
|
|
result.append(std::to_string(d.second));
|
|
|
|
}
|
|
|
|
else if (part == "ss")
|
|
|
|
{
|
|
|
|
if (d.second < 10)
|
|
|
|
{
|
|
|
|
result.append("0");
|
|
|
|
}
|
|
|
|
|
|
|
|
result.append(std::to_string(d.second));
|
|
|
|
}
|
|
|
|
else if (part == "am/pm" || part == "a/p")
|
|
|
|
{
|
|
|
|
if (d.hour < 12)
|
|
|
|
{
|
|
|
|
result.append("AM");
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
result.append("PM");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
index += part.size();
|
|
|
|
prev = index;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (index < format.value.size())
|
|
|
|
{
|
|
|
|
result.append(format.value.substr(index));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (format.value == "General" || format.value == "0")
|
|
|
|
{
|
|
|
|
if (number == static_cast<long long int>(number))
|
|
|
|
{
|
|
|
|
result = std::to_string(static_cast<long long int>(number));
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
result = std::to_string(number);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (format.value.substr(0, 8) == "#,##0.00" || format.value.substr(0, 9) == "-#,##0.00")
|
|
|
|
{
|
|
|
|
if (format.value[0] == '-')
|
|
|
|
{
|
|
|
|
result = "-";
|
|
|
|
}
|
|
|
|
|
|
|
|
if (format.has_locale && format.locale == "$$-1009")
|
|
|
|
{
|
|
|
|
result += "CA$";
|
|
|
|
}
|
|
|
|
else if (format.has_locale && format.locale == "$€-407")
|
|
|
|
{
|
|
|
|
result += "€";
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
result += "$";
|
|
|
|
}
|
|
|
|
|
|
|
|
result += std::to_string(number < 0 ? -number : number);
|
|
|
|
|
|
|
|
auto decimal_pos = result.find('.');
|
|
|
|
|
|
|
|
if (decimal_pos != std::string::npos)
|
|
|
|
{
|
|
|
|
result[decimal_pos] = ',';
|
|
|
|
decimal_pos += 3;
|
|
|
|
|
|
|
|
while (decimal_pos < result.size())
|
|
|
|
{
|
|
|
|
result.pop_back();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string format_section(const std::string &text, const section &format)
|
|
|
|
{
|
|
|
|
auto arobase_index = format.value.find('@');
|
|
|
|
|
|
|
|
std::string first_part, middle_part, last_part;
|
|
|
|
|
|
|
|
if (arobase_index != std::string::npos)
|
|
|
|
{
|
|
|
|
first_part = format.value.substr(0, arobase_index);
|
|
|
|
middle_part = text;
|
|
|
|
last_part = format.value.substr(arobase_index + 1);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
first_part = format.value;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto unquote = [](std::string &s) {
|
|
|
|
if (!s.empty())
|
|
|
|
{
|
|
|
|
if (s.front() != '"' || s.back() != '"') return false;
|
|
|
|
s = s.substr(0, s.size() - 2);
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
if (!unquote(first_part) || !unquote(last_part))
|
|
|
|
{
|
|
|
|
throw std::runtime_error(std::string("additional text must be enclosed in quotes: ") + format.value);
|
|
|
|
}
|
|
|
|
|
|
|
|
return first_part + middle_part + last_part;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string format_number(long double number, const std::string &format, xlnt::calendar base_date)
|
|
|
|
{
|
|
|
|
auto sections = parse_format_sections(format);
|
|
|
|
|
|
|
|
if (number > 0)
|
|
|
|
{
|
|
|
|
return format_section(number, sections.first, base_date);
|
|
|
|
}
|
|
|
|
else if (number < 0)
|
|
|
|
{
|
|
|
|
return format_section(number, sections.second, base_date);
|
|
|
|
}
|
|
|
|
|
|
|
|
// number == 0
|
|
|
|
return format_section(number, sections.third, base_date);
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string format_text(const std::string &text, const std::string &format)
|
|
|
|
{
|
|
|
|
if (format == "General") return text;
|
|
|
|
auto sections = parse_format_sections(format);
|
|
|
|
return format_section(text, sections.fourth);
|
|
|
|
}
|
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
} // namespace
|
2014-08-02 04:46:54 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
namespace xlnt {
|
2014-08-02 04:46:54 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::general()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(0), 0);
|
|
|
|
return *format;
|
2015-10-07 00:31:49 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::text()
|
2015-10-07 00:31:49 +08:00
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(49), 49);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::number()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(1), 1);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::number_00()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(2), 2);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::number_comma_separated1()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(4), 4);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::number_comma_separated2()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("#,##0.00_-");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::percentage()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(9), 9);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::percentage_00()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(10), 10);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_yyyymmdd2()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("yyyy-mm-dd");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_yyyymmdd()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("yy-mm-dd");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_ddmmyyyy()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("dd/mm/yy");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_dmyslash()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("d/m/y");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_dmyminus()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("d-m-y");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_dmminus()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("d-m");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_myminus()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("m-y");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_xlsx14()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(14), 14);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_xlsx15()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(15), 15);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_xlsx16()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(16), 16);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_xlsx17()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(17), 17);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_xlsx22()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(22), 22);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_datetime()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("yyyy-mm-dd h:mm:ss");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time1()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(18), 18);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time2()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(19), 19);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time3()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(20), 20);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time4()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(21), 21);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time5()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(45), 45);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time6()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format(builtin_formats().at(21), 21);
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time7()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("i:s.S");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_time8()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("h:mm:ss@");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_timedelta()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("[hh]:mm:ss");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::date_yyyymmddslash()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("yy/mm/dd@");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::currency_usd_simple()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("\"$\"#,##0.00_-");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::currency_usd()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("$#,##0_-");
|
|
|
|
return *format;
|
2015-10-24 02:42:36 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
const number_format number_format::currency_eur_simple()
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
static const number_format *format = new number_format("[$EUR ]#,##0.00_-");
|
|
|
|
return *format;
|
2014-08-02 04:46:54 +08:00
|
|
|
}
|
2015-10-19 03:30:46 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
number_format::number_format() : number_format(general())
|
2015-10-19 03:30:46 +08:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
2015-11-03 05:45:05 +08:00
|
|
|
number_format::number_format(std::size_t id) : number_format(from_builtin_id(id))
|
2015-10-19 03:30:46 +08:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
number_format::number_format(const std::string &format_string) : id_set_(false), id_(0)
|
2015-11-03 05:45:05 +08:00
|
|
|
{
|
|
|
|
set_format_string(format_string);
|
|
|
|
}
|
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
number_format::number_format(const std::string &format_string, std::size_t id) : id_set_(false), id_(0)
|
2015-10-19 03:30:46 +08:00
|
|
|
{
|
2015-10-24 02:42:36 +08:00
|
|
|
set_format_string(format_string, id);
|
2015-10-19 03:30:46 +08:00
|
|
|
}
|
|
|
|
|
2015-11-03 05:45:05 +08:00
|
|
|
number_format number_format::from_builtin_id(std::size_t builtin_id)
|
2014-07-24 08:51:28 +08:00
|
|
|
{
|
2015-11-01 22:43:01 +08:00
|
|
|
if (builtin_formats().find(builtin_id) == builtin_formats().end())
|
2014-07-24 08:51:28 +08:00
|
|
|
{
|
2015-11-11 07:58:54 +08:00
|
|
|
throw std::runtime_error("unknown id: " + std::to_string(builtin_id));
|
2014-07-24 08:51:28 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
auto format_string = builtin_formats().at(builtin_id);
|
|
|
|
return number_format(format_string, builtin_id);
|
2014-07-24 08:51:28 +08:00
|
|
|
}
|
2014-07-24 04:00:09 +08:00
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
std::string number_format::get_format_string() const
|
2015-10-19 03:30:46 +08:00
|
|
|
{
|
|
|
|
return format_string_;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::size_t number_format::hash() const
|
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
std::size_t seed = id_;
|
2015-10-24 02:42:36 +08:00
|
|
|
hash_combine(seed, format_string_);
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-10-24 02:42:36 +08:00
|
|
|
return seed;
|
2015-10-19 03:30:46 +08:00
|
|
|
}
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-11-03 05:45:05 +08:00
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
void number_format::set_format_string(const std::string &format_string)
|
2015-10-07 00:31:49 +08:00
|
|
|
{
|
2015-10-19 03:30:46 +08:00
|
|
|
format_string_ = format_string;
|
2015-11-03 05:45:05 +08:00
|
|
|
id_ = 0;
|
|
|
|
id_set_ = false;
|
2015-11-01 22:43:01 +08:00
|
|
|
|
2015-11-03 05:45:05 +08:00
|
|
|
for (const auto &pair : builtin_formats())
|
2015-10-07 00:31:49 +08:00
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
if (pair.second == format_string)
|
2015-10-19 03:30:46 +08:00
|
|
|
{
|
2015-11-03 05:45:05 +08:00
|
|
|
id_ = pair.first;
|
|
|
|
id_set_ = true;
|
|
|
|
break;
|
2015-10-19 03:30:46 +08:00
|
|
|
}
|
2015-10-07 00:31:49 +08:00
|
|
|
}
|
2015-10-19 03:30:46 +08:00
|
|
|
}
|
|
|
|
|
2015-11-11 07:58:54 +08:00
|
|
|
void number_format::set_format_string(const std::string &format_string, std::size_t id)
|
2015-11-03 05:45:05 +08:00
|
|
|
{
|
|
|
|
format_string_ = format_string;
|
|
|
|
id_ = id;
|
|
|
|
id_set_ = true;
|
|
|
|
}
|
|
|
|
|
2015-11-21 09:41:32 +08:00
|
|
|
bool number_format::has_id() const
|
|
|
|
{
|
|
|
|
return id_set_;
|
|
|
|
}
|
|
|
|
|
|
|
|
void number_format::set_id(std::size_t id)
|
|
|
|
{
|
|
|
|
id_ = id;
|
|
|
|
id_set_ = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::size_t number_format::get_id() const
|
|
|
|
{
|
|
|
|
if(!id_set_)
|
|
|
|
{
|
|
|
|
throw std::runtime_error("number format doesn't have an id");
|
|
|
|
}
|
|
|
|
|
|
|
|
return id_;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool number_format::is_date_format() const
|
|
|
|
{
|
|
|
|
if (format_string_ != "General")
|
|
|
|
{
|
|
|
|
try
|
|
|
|
{
|
|
|
|
auto sections = parse_format_sections(format_string_);
|
|
|
|
return ::is_date_format(sections.first.value);
|
|
|
|
}
|
|
|
|
catch (std::exception)
|
|
|
|
{
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string number_format::format(const std::string &text) const
|
|
|
|
{
|
|
|
|
return format_text(text, format_string_);
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string number_format::format(long double number, calendar base_date) const
|
|
|
|
{
|
|
|
|
return format_number(number, format_string_, base_date);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool number_format::operator==(const number_format &other) const
|
|
|
|
{
|
|
|
|
return hash() == other.hash();
|
|
|
|
}
|
|
|
|
|
2014-07-24 04:00:09 +08:00
|
|
|
} // namespace xlnt
|