1
0
mirror of https://github.com/zenorogue/hyperrogue.git synced 2024-12-20 23:50:27 +00:00
hyperrogue/langen.cpp

474 lines
12 KiB
C++

// Hyperbolic Rogue language file generator
// Copyright (C) 2011-2018 Zeno Rogue, see 'hyper.cpp' for details
#include <map>
#include <string>
#include <cstdio>
#include <vector>
#include <cstdlib>
#include <set>
#define GEN_M 0
#define GEN_F 1
#define GEN_N 2
#define GEN_O 3
#define GENF_ELISION (1 << 3)
#define GENF_PLURALONLY (1 << 4)
#define GENF_PROPER (1 << 5)
#if MAC
#define IF_MAC(y,z) y
#else
#define IF_MAC(y,z) z
#endif
template<class T> int isize(const T& x) { return x.size(); }
#define NUMLAN 8
// language generator
std::string current_language;
const char *escape(std::string s, const std::string& dft);
template<class T> struct dictionary {
std::map<std::string, T> m;
void add(const std::string& s, T val) {
auto it = m.find(s);
if (it == m.end()) {
m.emplace(s, std::move(val));
}
else if (val != it->second) {
printf("// #warning Two translations for %s [%s]\n", escape(s, s), current_language.c_str());
}
}
T& operator [] (const std::string& s) { return m[s]; }
int count(const std::string& s) const { return m.count(s); }
};
dictionary<std::string> d[NUMLAN];
struct noun2 {
int genus;
const char *nom;
const char *nomp;
const char *acc;
const char *abl;
};
struct noun {
int genus;
std::string nom, nomp, acc, abl;
noun() = default;
noun(const noun2& n) : genus(n.genus), nom(n.nom), nomp(n.nomp), acc(n.acc), abl(n.abl) {}
friend bool operator==(const noun& a, const noun& b) {
return std::tie(a.genus, a.nom, a.nomp, a.acc, a.abl) == std::tie(b.genus, b.nom, b.nomp, b.acc, b.abl);
}
friend bool operator!=(const noun& a, const noun& b) {
return std::tie(a.genus, a.nom, a.nomp, a.acc, a.abl) != std::tie(b.genus, b.nom, b.nomp, b.acc, b.abl);
}
};
dictionary<noun> nouns[NUMLAN];
int utfsize(char c) {
unsigned char cu = c;
if(cu < 128) return 1;
if(cu < 224) return 2;
if(cu < 0xF0) return 3;
return 4;
}
void addutftoset(std::set<std::string>& s, const std::string& w) {
size_t i = 0;
while(i < w.size()) {
int siz = utfsize(w[i]);
s.insert(w.substr(i, siz));
i += siz;
}
}
void addutftoset(std::set<std::string>& s, const noun& w) {
addutftoset(s, w.nom);
addutftoset(s, w.nomp);
addutftoset(s, w.acc);
addutftoset(s, w.abl);
}
template<class T>
void addutftoset(std::set<std::string>& s, const dictionary<T>& w) {
for(auto&& elt : w.m)
addutftoset(s, elt.second);
}
std::set<std::string> allchars;
typedef unsigned hashcode;
hashcode hashval;
hashcode langhash(const std::string& s) {
hashcode r = 0;
for (char ch : s) r = hashval * r + ch;
return r;
}
std::map<hashcode, std::string> buildHashTable(std::set<std::string>& s) {
std::map<hashcode, std::string> res;
for(auto&& elt : s)
res[langhash(elt)] = elt;
return res;
}
const char *escape(std::string s, const std::string& dft) {
if(s == "") {
printf("/*MISSING*/ ");
s = dft;
}
static std::string t;
t = "\"";
for(int i=0; i<isize(s); i++)
if(s[i] == '\\') t += "\\\\";
else if(s[i] == '\n') t += "\\n";
else if(s[i] == '\"') t += "\\\"";
else t += s[i];
t += "\"";
return t.c_str();
}
std::set<std::string> nothe;
std::set<std::string> plural;
void langPL() {
current_language = "PL";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e,f)
#include "language-pl.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e,f) { a, noun2{ b, c, d, e, f } },
#include "language-pl.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[1].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[1].add(elt.first, elt.second);
current_language = "-";
}
void langTR() {
current_language = "TR";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e,f)
#include "language-tr.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e,f) { a, noun2{ b, c, d, e, f } },
#include "language-tr.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[2].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[2].add(elt.first, elt.second);
current_language = "-";
}
void langCZ() {
current_language = "CZ";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e,f)
#include "language-cz.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e,f) { a, noun2{ b, c, d, e, f } },
#include "language-cz.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[3].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[3].add(elt.first, elt.second);
current_language = "-";
}
void langRU() {
current_language = "RU";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e,f)
#include "language-ru.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e,f) { a, noun2{ b, c, d, e, f } },
#include "language-ru.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[4].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[4].add(elt.first, elt.second);
current_language = "-";
}
void langDE() {
current_language = "DE";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e)
#include "language-de.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e) { a, noun2{ b, c, d, e, e } },
#include "language-de.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[5].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[5].add(elt.first, elt.second);
current_language = "-";
}
void langPT() {
current_language = "PT";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e)
#include "language-ptbr.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e) { a, noun2{ b, c, d, "", e } },
#include "language-ptbr.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[6].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[6].add(elt.first, elt.second);
current_language = "-";
}
void langFR() {
current_language = "FR";
static std::pair<const char *, const char *> ds[] = {
#define S(a,b) { a, b },
#define N(a,b,c,d,e)
#include "language-fr.cpp"
#undef N
#undef S
};
static std::pair<const char *, noun2> ns[] = {
#define S(a,b)
#define N(a,b,c,d,e) { a, noun2{ b, c, d, e, e } },
#include "language-fr.cpp"
#undef N
#undef S
};
for(auto&& elt : ds) d[7].add(elt.first, elt.second);
for(auto&& elt : ns) nouns[7].add(elt.first, elt.second);
current_language = "-";
}
int completeness[NUMLAN];
template<class T>
void compute_completeness(const T& dict)
{
std::set<std::string> s;
for(int i=1; i<NUMLAN; i++)
for(auto&& elt : dict[i].m)
s.insert(elt.first);
for(auto&& elt : s) {
std::string mis = "", mis1 = "", exist_in = "";
bool in_important = false;
for(int i=1; i<NUMLAN; i++) if(dict[i].count(elt) == 0) {
std::string which = d[i]["EN"];
if(which != "TR" && which != "DE" && which != "PT-BR" && which != "RU")
mis += which + " ";
else
mis1 += which + " ";
}
else {
std::string which = d[i]["EN"];
if(which != "PT-BR" && which != "TR" && which != "DE")
in_important = true;
exist_in += which + " ";
}
if(mis != "") mis.pop_back();
if(mis1 != "") mis1.pop_back();
if(exist_in != "") exist_in.pop_back();
if(in_important && mis != "")
printf("// #warning Missing [%s : %s] from [%s]: %s\n", mis.c_str(), mis1.c_str(), exist_in.c_str(), escape(elt, "?"));
completeness[0]++;
for(int i=1; i<NUMLAN; i++) if(dict[i].count(elt)) completeness[i]++;
}
}
int main() {
printf("// DO NOT EDIT -- this file is generated automatically with langen\n\n");
nothe.insert("R'Lyeh");
nothe.insert("Camelot");
nothe.insert("Hell");
plural.insert("Crossroads");
plural.insert("Crossroads II");
plural.insert("Crossroads III");
plural.insert("Elemental Planes");
plural.insert("Crossroads IV");
plural.insert("Kraken Depths");
allchars.insert("");
allchars.insert("δ");
allchars.insert("");
allchars.insert("½");
allchars.insert("²");
allchars.insert("π");
allchars.insert("Θ");
allchars.insert("λ");
allchars.insert("");
allchars.insert("");
allchars.insert("");
allchars.insert("");
allchars.insert("");
allchars.insert("");
langPL(); langCZ(); langRU();
langTR(); langDE(); langPT();
langFR();
// verify
compute_completeness(d);
compute_completeness(nouns);
for(int i=1; i<NUMLAN; i++) {
addutftoset(allchars, d[i]);
addutftoset(allchars, nouns[i]);
}
std::string javastring;
std::vector<std::string> vchars;
for(auto&& elt : allchars) {
if(isize(elt) >= 2) { javastring += elt; vchars.push_back(elt); }
}
printf("\n");
printf("#if HDR\n");
printf("#if CAP_TRANS\n");
printf("#define NUMEXTRA %d\n", isize(vchars));
printf("#define NATCHARS {");
for(auto&& elt : vchars) printf("\"%s\",", elt.c_str());
printf("}\n");
printf("extern const char* natchars[NUMEXTRA];\n");
printf("#endif\n");
printf("#endif\n");
printf("const char* natchars[NUMEXTRA] = NATCHARS;\n");
printf("//javastring = \"%s\";\n", javastring.c_str());
printf("\nEX int transcompleteness[NUMLAN] = {");
for(int i=0; i<NUMLAN; i++) printf("%d, ", completeness[i]);
printf("};\n");
printf("\n//statistics\n");
for(auto&& elt : d[1].m)
d[0][elt.first] = elt.first;
for(auto&& elt : nouns[1].m) {
noun n = elt.second;
n.nom = n.nomp = n.acc = n.abl = elt.first;
nouns[0][elt.first] = n;
}
printf("// total: %5d nouns, %5d sentences\n", isize(nouns[1].m), isize(d[1].m));
for(int i=0; i<NUMLAN; i++) {
size_t bnouns = 0;
size_t bdict = 0;
for(auto&& elt : d[i].m)
bdict += elt.second.size();
for(auto&& elt : nouns[i].m) {
const noun& n = elt.second;
bnouns += n.nom.size();
bnouns += n.nomp.size();
bnouns += n.acc.size();
bnouns += n.abl.size();
}
printf("// %s: %5dB nouns, %5dB sentences\n",
d[i]["EN"].c_str(), int(bnouns), int(bdict));
}
std::set<std::string> allsent;
for(auto&& elt : d[1].m)
allsent.insert(elt.first);
std::set<std::string> allnouns;
for(auto&& elt : nouns[1].m)
allnouns.insert(elt.first);
std::map<hashcode, std::string> ms, mn;
do {
hashval = rand();
printf("// check hash: %x\n", hashval);
ms = buildHashTable(allsent);
mn = buildHashTable(allnouns);
}
while(ms.size() != allsent.size() || mn.size() != allnouns.size());
printf("hashcode hashval = 0x%x;\n\n", hashval);
printf("sentence all_sentences[] = {\n");
for(auto&& elt : ms) {
const std::string& s = elt.second;
printf(" {0x%x, { // %s\n", elt.first, escape(s, s));
for(int i=1; i<NUMLAN; i++) printf(" %s,\n", escape(d[i][s], s));
printf(" }},\n");
}
printf(" };\n\n");
printf("fullnoun all_nouns[] = {\n");
for(auto&& elt : mn) {
const std::string& s = elt.second;
printf(" {0x%x, %d, { // \"%s\"\n", elt.first,
(nothe.count(s) ? 1:0) + (plural.count(s) ? 2:0),
escape(s, s));
for(int i=1; i<NUMLAN; i++) {
printf(" {%d", nouns[i][s].genus);
printf(", %s", escape(nouns[i][s].nom, s));
printf(", %s", escape(nouns[i][s].nomp, s));
printf(", %s", escape(nouns[i][s].acc, s));
printf(", %s},\n", escape(nouns[i][s].abl, s));
}
printf(" }},\n");
}
printf(" };\n");
}