|
DataMuseum.dkPresents historical artifacts from the history of: DKUUG/EUUG Conference tapes |
This is an automatic "excavation" of a thematic subset of
See our Wiki for more about DKUUG/EUUG Conference tapes Excavated with: AutoArchaeologist - Free & Open Source Software. |
top - metrics - downloadIndex: T k
Length: 29450 (0x730a) Types: TextFile Names: »key-list.cc«
└─⟦a05ed705a⟧ Bits:30007078 DKUUG GNU 2/12/89 └─⟦cc8755de2⟧ »./libg++-1.36.1.tar.Z« └─⟦23757c458⟧ └─⟦this⟧ »libg++/gperf/src/key-list.cc«
/* Routines for building, ordering, and printing the keyword list. Copyright (C) 1989 Free Software Foundation, Inc. written by Douglas C. Schmidt (schmidt@ics.uci.edu) This file is part of GNU GPERF. GNU GPERF is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 1, or (at your option) any later version. GNU GPERF is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with GNU GPERF; see the file COPYING. If not, write to the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */ #include <stdio.h> #include <builtin.h> #include <assert.h> #include "options.h" #include "read-line.h" #include "hash-table.h" #include "key-list.h" /* Current release version. */ extern char *version_string; /* How wide the printed field width must be to contain the maximum hash value. */ static int field_width = 2; /* Destructor dumps diagnostics during debugging. */ Key_List::~Key_List (void) { if (option[DEBUG]) { report_error ("\nDumping key list information:\ntotal unique keywords = %d" "\ntotal keywords = %d\nmaximum key length = %d.\n", list_len, total_keys, max_key_len); dump (); report_error ("End dumping list.\n\n"); } } /* Gathers the input stream into a buffer until one of two things occur: 1. We read a '%' followed by a '%' 2. We read a '%' followed by a '}' The first symbolizes the beginning of the keyword list proper, The second symbolizes the end of the C source code to be generated verbatim in the output file. I assume that the keys are separated from the optional preceding struct declaration by a consecutive % followed by either % or } starting in the first column. The code below uses an expandible buffer to scan off and return a pointer to all the code (if any) appearing before the delimiter. */ char * Key_List::get_special_input (char delimiter) { int size = 80; char *buf = new char[size]; int c, i; for (i = 0; (c = getchar ()) != EOF; i++) { if (c == '%') { if ((c = getchar ()) == delimiter) { while ((c = getchar ()) != '\n') ; /* discard newline */ if (i == 0) return ""; else { buf[delimiter == '%' && buf[i - 2] == ';' ? i - 2 : i - 1] = '\0'; return buf; } } else ungetc (c, stdin); } else if (i >= size) /* Yikes, time to grow the buffer! */ { char *temp = new char[size *= 2]; int j; for (j = 0; j < i; j++) temp[j] = buf[j]; delete (buf); buf = temp; } buf[i] = c; } return NULL; /* Problem here. */ } /* Stores any C text that must be included verbatim into the generated code output. */ char * Key_List::save_include_src (void) { int c; if ((c = getchar ()) != '%') ungetc (c, stdin); else if ((c = getchar ()) != '{') report_error ("internal error, %c != '{' on line %d in file %s%a", c, __LINE__, __FILE__); else return get_special_input ('}'); return ""; } /* Determines from the input file whether the user wants to build a table from a user-defined struct, or whether the user is content to simply use the default array of keys. */ char * Key_List::get_array_type (void) { return get_special_input ('%'); } /* strcspn - find length of initial segment of S consisting entirely of characters not from REJECT (borrowed from Henry Spencer's ANSI string package, when GNU libc comes out I'll replace this...). */ static inline int Key_List::strcspn (char *s, char *reject) { char *scan; char *rej_scan; int count = 0; for (scan = s; *scan; scan++) { for (rej_scan = reject; *rej_scan; rej_scan++) if (*scan == *rej_scan) return count; count++; } return count; } /* Sets up the Return_Type, the Struct_Tag type and the Array_Type based upon various user Options. */ void Key_List::set_output_types (void) { if (option[TYPE] && !(array_type = get_array_type ())) return; /* Something's wrong, bug we'll catch it later on.... */ else if (option[TYPE]) /* Yow, we've got a user-defined type... */ { int struct_tag_length = strcspn (array_type, "{\n\0"); if (option[POINTER]) /* And it must return a pointer... */ { return_type = new char[struct_tag_length + 2]; strncpy (return_type, array_type, struct_tag_length); return_type[struct_tag_length] = '\0'; strcat (return_type, "*"); } struct_tag = new char[struct_tag_length + 1]; strncpy (struct_tag, array_type, struct_tag_length); struct_tag[struct_tag_length] = '\0'; } else if (option[POINTER]) /* Return a char *. */ return_type = default_array_type; } /* Reads in all keys from standard input and creates a linked list pointed to by Head. This list is then quickly checked for ``links,'' i.e., unhashable elements possessing identical key sets and lengths. */ void Key_List::read_keys (void) { char *ptr; include_src = save_include_src (); set_output_types (); /* Oops, problem with the input file. */ if (! (ptr = Read_Line::get_line ())) report_error ("No words in input file, did you forget to prepend %s" " or use -t accidentally?\n%a", "%%"); /* Read in all the keywords from the input file. */ else { List_Node *temp, *trail; head = new List_Node (ptr, strcspn (ptr, ",\n")); for (temp = head; (ptr = Read_Line::get_line ()) && strcmp (ptr, "%%"); temp = temp->next) { temp->next = new List_Node (ptr, strcspn (ptr, ",\n")); total_keys++; } /* See if any additional C code is included at end of this file. */ if (ptr) additional_code = 1; /* If this becomes TRUE we've got a link. */ int link = 0; /* Hash table this number of times larger than keyword number. */ const int table_multiple = 5; /* Make large hash table for efficiency. */ Hash_Table found_link ((list_len = total_keys) * table_multiple); /* Test whether there are any links and also set the maximum length of an identifier in the keyword list. */ for (temp = head, trail = NULL; temp; temp = temp->next) { List_Node *ptr = found_link (temp, option[NOLENGTH]); /* Check for links. We deal with these by building an equivalence class of all duplicate values (i.e., links) so that only 1 keyword is representative of the entire collection. This *greatly* simplifies processing during later stages of the program. */ if (ptr) { list_len--; trail->next = temp->next; temp->link = ptr->link; ptr->link = temp; link = 1; /* Complain if user hasn't enabled the duplicate option. */ if (!option[DUP]) report_error ("Key link: \"%s\" = \"%s\", with key set \"%s\".\n", temp->key, ptr->key, temp->key_set); else if (option[DEBUG]) report_error ("Key link: \"%s\" = \"%s\", with key set \"%s\".\n", temp->key, ptr->key, temp->key_set); } else trail = temp; /* Update minimum and maximum keyword length, if needed. */ max_key_len >?= temp->length; min_key_len <?= temp->length; } /* Exit program if links exists and option[DUP] not set, since we can't continue */ if (link) { if (option[DUP]) { if (!option[SWITCH]) { report_error ("turning on -S option.\n"); option = SWITCH; } report_error ("Some input keys have identical hash values, examine output carefully...\n"); } else report_error ("Some input keys have identical hash values,\ntry different key positions or use option -D.\n%a"); } /* If no links, clear the DUP option so we can use the length table, if output. */ else if (option[DUP]) option != DUP; } } /* Recursively merges two sorted lists together to form one sorted list. The ordering criteria is by frequency of occurrence of elements in the key set or by the hash value. This is a kludge, but permits nice sharing of almost identical code without incurring the overhead of a function call comparison. */ List_Node * Key_List::merge (List_Node *list1, List_Node *list2) { if (!list1) return list2; else if (!list2) return list1; else if (occurrence_sort && list1->occurrence < list2->occurrence || hash_sort && list1->hash_value > list2->hash_value) { list2->next = merge (list2->next, list1); return list2; } else { list1->next = merge (list1->next, list2); return list1; } } /* Applies the merge sort algorithm to recursively sort the key list by frequency of occurrence of elements in the key set. */ List_Node * Key_List::merge_sort (List_Node *head) { if (!head || !head->next) return head; else { List_Node *middle = head; List_Node *temp = head->next->next; while (temp) { temp = temp->next; middle = middle->next; if (temp) temp = temp->next; } temp = middle->next; middle->next = NULL; return merge (merge_sort (head), merge_sort (temp)); } } /* Returns the frequency of occurrence of elements in the key set. */ static inline int Key_List::get_occurrence (List_Node *ptr) { int value = 0; char *temp; for (temp = ptr->key_set; *temp; temp++) value += occurrences[*temp]; return value; } /* Enables the index location of all key set elements that are now determined. */ static inline void Key_List::set_determined (List_Node *ptr) { char *temp; for (temp = ptr->key_set; *temp; temp++) determined[*temp] = 1; } /* Returns TRUE if PTR's key set is already completely determined. */ static inline int Key_List::already_determined (List_Node *ptr) { int is_determined = 1; char *temp; for (temp = ptr->key_set; is_determined && *temp; temp++) is_determined = determined[*temp]; return is_determined; } /* Reorders the table by first sorting the list so that frequently occuring keys appear first, and then the list is reorded so that keys whose values are already determined will be placed towards the front of the list. This helps prune the search time by handling inevitable collisions early in the search process. See Cichelli's paper from Jan 1980 JACM for details.... */ void Key_List::reorder (void) { List_Node *ptr; for (ptr = head; ptr; ptr = ptr->next) ptr->occurrence = get_occurrence (ptr); hash_sort = 0; occurrence_sort = 1; for (ptr = head = merge_sort (head); ptr->next; ptr = ptr->next) { set_determined (ptr); if (already_determined (ptr->next)) continue; else { List_Node *trail_ptr = ptr->next; List_Node *run_ptr = trail_ptr->next; for (; run_ptr; run_ptr = trail_ptr->next) { if (already_determined (run_ptr)) { trail_ptr->next = run_ptr->next; run_ptr->next = ptr->next; ptr = ptr->next = run_ptr; } else trail_ptr = run_ptr; } } } } /* Determines the maximum and minimum hash values. One notable feature is Ira Pohl's optimal algorithm to calculate both the maximum and minimum items in a list in O(3n/2) time (faster than the O (2n) method). Returns the maximum hash value encountered. */ int Key_List::print_min_max (void) { int min_hash_value; int max_hash_value; List_Node *temp; if (odd (list_len)) /* Pre-process first item, list now has an even length. */ { min_hash_value = max_hash_value = head->hash_value; temp = head->next; } else /* List is already even length, no problem extra work necessary. */ { min_hash_value = MAX_INT; max_hash_value = NEG_MAX_INT; temp = head; } for ( ; temp; temp = temp->next) /* Find max and min in optimal O(3n/2) time. */ { int key_2, key_1 = temp->hash_value; temp = temp->next; key_2 = temp->hash_value; if (key_1 < key_2) { min_hash_value <?= key_1; max_hash_value >?= key_2; } else { min_hash_value <?= key_2; max_hash_value >?= key_1; } } printf ("\n#define MIN_WORD_LENGTH %d\n#define MAX_WORD_LENGTH %d" "\n#define MIN_HASH_VALUE %d\n#define MAX_HASH_VALUE %d" "\n/*\n%5d keywords\n%5d is the maximum key range\n*/\n\n", min_key_len == MAX_INT ? max_key_len : min_key_len, max_key_len, min_hash_value, max_hash_value, total_keys, (max_hash_value - min_hash_value + 1)); return max_hash_value; } /* Generates the output using a C switch. This trades increased search time for decreased table space (potentially *much* less space for sparse tables). It the user has specified their own struct in the keyword file *and* they enable the POINTER option we have extra work to do. The solution here is to maintain a local static array of user defined struct's, as with the Print_Lookup_Function. Then we use for switch statements to perform either a strcmp or strncmp, returning 0 if the str fails to match, and otherwise returning a pointer to appropriate index location in the local static array. */ #ifdef sparc #include <alloca.h> #endif void Key_List::print_switch (void) { char *comp_buffer; int pointer_and_type_enabled = option[POINTER] && option[TYPE]; if (pointer_and_type_enabled) { comp_buffer = (char *) alloca (strlen ("*str == *resword->%s && !strncmp (str + 1, resword->%s + 1, len - 1)" + 2 * strlen (option.get_key_name ()) + 1)); sprintf (comp_buffer, option[COMP] ? "*str == *resword->%s && !strncmp (str + 1, resword->%s + 1, len - 1)" : "*str == *resword->%s && !strcmp (str + 1, resword->%s + 1)", option.get_key_name (), option.get_key_name ()); } else comp_buffer = option[COMP] ? "*str == *resword && !strncmp (str + 1, resword + 1, len - 1)" : "*str == *resword && !strcmp (str + 1, resword + 1)"; printf (" if (len <= MAX_WORD_LENGTH && len >= MIN_WORD_LENGTH)\n {\n" " register int key = %s (str, len);\n\n" " if (key <= MAX_HASH_VALUE && key >= MIN_HASH_VALUE)\n {\n", option.get_hash_name ()); /* Output each keyword as part of a switch statement indexed by hash value. */ if (option[POINTER] || option[DUP]) { List_Node *temp; printf (" %s *resword; %s\n\n switch (key)\n {\n", pointer_and_type_enabled ? struct_tag : "char", option[LENTABLE] && !option[DUP] ? "int key_len;" : ""); for (temp = head; temp; temp = temp->next) { printf (" case %*d:\n", field_width, temp->hash_value); if (temp->link) { List_Node *links; for (links = temp; links; links = links->link) if (pointer_and_type_enabled) printf (" resword = &wordlist[%d];\n if (%s) return resword;\n", links->index, comp_buffer); else printf (" resword = \"%s\";\n if (%s) return resword;\n", links->key, comp_buffer); printf (" return 0;\n"); } else if (temp->next && temp->hash_value == temp->next->hash_value) { for ( ; temp->next && temp->hash_value == temp->next->hash_value; temp = temp->next) if (pointer_and_type_enabled) printf (" resword = &wordlist[%d];\n if (%s) return resword;\n", temp->index, comp_buffer); else printf (" resword = \"%s\";\n if (%s) return resword;\n", temp->key, comp_buffer); if (pointer_and_type_enabled) printf (" resword = &wordlist[%d];\n return %s ? resword : 0;\n", temp->index, comp_buffer); else printf (" resword = \"%s\";\n return %s ? resword : 0;\n", temp->key, comp_buffer); } else { if (pointer_and_type_enabled) printf (" resword = &wordlist[%d];", temp->index); else printf (" resword = \"%s\";", temp->key); if (option[LENTABLE] && !option[DUP]) printf (" key_len = %d;", temp->length); printf (" break;\n"); } } printf (" default: return 0;\n }\n"); printf (option[LENTABLE] && !option[DUP] ? " if (len == key_len && %s)\n return resword;\n" : " if (%s)\n return resword;\n", comp_buffer); printf (" }\n }\n return 0;\n}\n"); } else /* Nothing special required here. */ { List_Node *temp; printf (" char *s = \"\";\n\n switch (key)\n {\n"); for (temp = head; temp; temp = temp->next) if (option[LENTABLE]) printf (" case %*d: if (len == %d) s = \"%s\"; else return 0; break;\n", field_width, temp->hash_value, temp->length, temp->key); else printf (" case %*d: s = \"%s\"; break;\n", field_width, temp->hash_value, temp->key); printf (" default: return 0;\n }\n return *s == *str && !%s;\n }\n }\n return 0;\n}\n", option[COMP] ? "strncmp (s + 1, str + 1, len - 1)" : "strcmp (s + 1, str + 1)"); } } /* Prints out a table of keyword lengths, for use with the comparison code in generated function ``in_word_set.'' */ void Key_List::print_keylength_table (void) { const int max_column = 15; int index = 0; int column = 0; char *indent = option[GLOBAL] ? "" : " "; List_Node *temp; if (!option[DUP] && !option[SWITCH]) { printf ("\n%sstatic %sunsigned %s lengthtable[] =\n%s%s{\n ", indent, option[CONST] ? "const " : "", max_key_len < 256 ? "char" : (max_key_len < 65536 ? "short" : "long"), indent, indent); for (temp = head; temp; temp = temp->next, index++) { if (index < temp->hash_value) for ( ; index < temp->hash_value; index++) printf ("%3d%s", 0, ++column % (max_column - 1) ? "," : ",\n "); printf ("%3d%s", temp->length, ++column % (max_column - 1 ) ? "," : ",\n "); } printf ("\n%s%s};\n", indent, indent); } } /* Prints out the array containing the key words for the Gen_Perf hash function. */ void Key_List::print_keyword_table (void) { char *l_brace = *head->rest ? "{" : ""; char *r_brace = *head->rest ? "}," : ""; int doing_switch = option[SWITCH]; char *indent = option[GLOBAL] ? "" : " "; int index = 0; List_Node *temp; printf ("\n%sstatic %s%s wordlist[] =\n%s%s{\n", indent, option[CONST] ? "const " : "", struct_tag, indent, indent); /* generate an array of reserved words at appropriate locations */ for (temp = head; temp; temp = temp->next, index++) { temp->index = index; if (!doing_switch && index < temp->hash_value) { int column; printf (" "); for (column = 1; index < temp->hash_value; index++, column++) printf ("%s\"\",%s %s", l_brace, r_brace, column % 9 ? "" : "\n "); if (column % 10) printf ("\n"); else { printf ("%s\"%s\", %s%s\n", l_brace, temp->key, temp->rest, r_brace); continue; } } printf (" %s\"%s\", %s%s\n", l_brace, temp->key, temp->rest, r_brace); /* Deal with links specially. */ if (temp->link) { List_Node *links; for (links = temp->link; links; links = links->link) { links->index = ++index; printf (" %s\"%s\", %s%s\n", l_brace, links->key, links->rest, r_brace); } } } printf ("%s%s};\n\n", indent, indent); } /* Generates C code for the hash function that returns the proper encoding for each key word. */ void Key_List::print_hash_function (int max_hash_value) { const int max_column = 10; int count = max_hash_value; /* Calculate maximum number of digits required for MAX_HASH_VALUE. */ while ((count /= 10) > 0) field_width++; if (option[GNU]) printf ("#ifdef __GNUC__\ninline\n#endif\n"); printf (option[ANSI] ? "static int\n%s (register const char *str, register int len)\n{\n static %sunsigned %s asso_value[] =\n {" : "static int\n%s (str, len)\n register char *str;\n register int unsigned len;\n{\n static %sunsigned %s asso_value[] =\n {", option.get_hash_name (), option[CONST] ? "const " : "", max_hash_value < 256 ? "char" : (max_hash_value < 65536 ? "short" : "int")); for (count = 0; count < ALPHA_SIZE; ++count) { if (!(count % max_column)) printf ("\n "); printf ("%*d,", field_width, occurrences[count] ? asso_values[count] : max_hash_value + 1); } /* Optimize special case of ``-k 1,$'' */ if (option[DEFAULTCHARS]) printf ("\n };\n return %sasso_value[str[len - 1]] + asso_value[str[0]];\n}\n\n", option[NOLENGTH] ? "" : "len + "); else { int key_pos; option.reset (); /* Get first (also highest) key position. */ key_pos = option.get (); /* We can perform additional optimizations here. */ if (!option[ALLCHARS] && key_pos <= min_key_len) { printf ("\n };\n return %s", option[NOLENGTH] ? "" : "len + "); for (; key_pos != WORD_END; ) { printf ("asso_value[str[%d]]", key_pos - 1); if ((key_pos = option.get ()) != EOS) printf (" + "); else break; } printf ("%s;\n}\n\n", key_pos == WORD_END ? "asso_value[str[len - 1]]" : ""); } /* We've got to use the correct, but brute force, technique. */ else { printf ("\n };\n register int hval = %s ;\n\n switch (%s)\n {\n default:\n", option[NOLENGTH] ? "0" : "len", option[NOLENGTH] ? "len" : "hval"); /* User wants *all* characters considered in hash. */ if (option[ALLCHARS]) { int i; for (i = max_key_len; i > 0; i--) printf (" case %d:\n hval += asso_value[str[%d]];\n", i, i - 1); printf (" }\n return hval;\n}\n\n"); } else /* do the hard part... */ { count = key_pos + 1; do { while (--count > key_pos) printf (" case %d:\n", count); printf (" case %d:\n hval += asso_value[str[%d]];\n", key_pos, key_pos - 1); } while ((key_pos = option.get ()) != EOS && key_pos != WORD_END); printf (" }\n return hval%s;\n}\n\n", key_pos == WORD_END ? " + asso_value[str[len - 1]]" : ""); } } } } /* Generates C code to perform the keyword lookup. */ void Key_List::print_lookup_function (void) { printf (" if (len <= MAX_WORD_LENGTH && len >= MIN_WORD_LENGTH)\n {\n" " register int key = %s (str, len);\n\n" " if (key <= MAX_HASH_VALUE && key >= MIN_HASH_VALUE)\n {\n" " register %schar *s = wordlist[key]", option.get_hash_name (), option[CONST] ? "const " : ""); if (array_type != default_array_type) printf (".%s", option.get_key_name ()); printf (";\n\n if (%s*s == *str && !%s)\n return %s", option[LENTABLE] ? "len == lengthtable[key]\n && " : "", option[COMP] ? "strncmp (str + 1, s + 1, len - 1)" : "strcmp (str + 1, s + 1)", option[TYPE] && option[POINTER] ? "&wordlist[key]" : "s"); printf (";\n }\n }\n return 0;\n}\n"); } /* Generates the hash function and the key word recognizer function based upon the user's Options. */ void Key_List::output (void) { int global_table = option[GLOBAL] ; printf ("/* C code produced by gperf version %s */\n", version_string); option.print_options (); printf ("%s\n", include_src); if (option[TYPE] && !option[NOTYPE]) /* Output type declaration now, reference it later on.... */ printf ("%s;\n", array_type); print_hash_function (print_min_max ()); if (global_table) if (option[SWITCH]) { if (option[LENTABLE] && option[DUP]) print_keylength_table (); if (option[POINTER] && option[TYPE]) print_keyword_table (); } else { if (option[LENTABLE]) print_keylength_table (); print_keyword_table (); } if (option[GNU]) /* Use the inline keyword to remove function overhead. */ printf ("#ifdef __GNUC__\ninline\n#endif\n"); printf (option[ANSI] ? "%s%s\n%s (register const char *str, register int len)\n{\n" : "%s%s\n%s (str, len)\n register char *str;\n register unsigned int len;\n{\n", option[CONST] ? "const " : "", return_type, option.get_function_name ()); /* Use the switch in place of lookup table. */ if (option[SWITCH]) { if (!global_table) { if (option[LENTABLE] && option[DUP]) print_keylength_table (); if (option[POINTER] && option[TYPE]) print_keyword_table (); } print_switch (); } else /* Use the lookup table, in place of switch. */ { if (!global_table) { if (option[LENTABLE]) print_keylength_table (); print_keyword_table (); } print_lookup_function (); } if (additional_code) { int c; while ((c = getchar ()) != EOF) putchar (c); } } /* Sorts the keys by hash value. */ void Key_List::sort (void) { hash_sort = 1; occurrence_sort = 0; head = merge_sort (head); } /* Dumps the key list to stderr stream. */ void Key_List::dump () { List_Node *ptr; report_error ("\nList contents are:\n" "(hash value, key length, index, key set, uniq set, key):\n"); for (ptr = head; ptr; ptr = ptr->next) report_error (" %d, %d, %d, %s, %s, %s\n", ptr->hash_value, ptr->length, ptr->index, ptr->key_set, ptr->uniq_set, ptr->key); } /* Simple-minded constructor action here... */ Key_List::Key_List (void) { total_keys = 1; max_key_len = NEG_MAX_INT; min_key_len = MAX_INT; return_type = default_return_type; array_type = struct_tag = default_array_type; head = NULL; additional_code = 0; } /* Returns the length of entire key list. */ int Key_List::length (void) { return list_len; } /* Returns length of longest key read. */ int Key_List::max_key_length (void) { return max_key_len; }