regex-engine/lib/construct.c

/*
 * Copyright (c) Camden Dixie O'Brien
 * SPDX-License-Identifier: AGPL-3.0-only
 */

#include "construct.h"

#include <assert.h>
#include <stdlib.h>
#include <string.h>

static void add_fsa(fsa_t *f, const fsa_t *o, int *init_out, int *final_out)
{
	assert(f != o);

	// Ensure f has enough space for o's states, then copy o's states
	// into f.
	const int count = f->count + o->count;
	if (f->capacity < count) {
		do
			f->capacity *= 2;
		while (f->capacity < count);
		f->states = realloc(f->states, f->capacity * sizeof(fsa_state_t));
		assert(f->states);
	}
	memcpy(f->states + f->count, o->states, o->count * sizeof(fsa_state_t));

	// Mark o's final state as non-final.
	f->states[f->count].final = false;

	// Retarget the rules of the copied states to refer to the new
	// state indices.
	for (int i = f->count; i < count; ++i) {
		for (int j = 0; j < f->states[i].count; ++j)
			f->states[i].rules[j].next += f->count;
	}

	// Clean up o's remaining resources.  All of the states have been
	// copied to f so we just need to free its states buffer.
	free(o->states);

	if (NULL != init_out)
		*init_out = o->initial + f->count;
	if (NULL != final_out)
		*final_out = f->count;
	f->count = count;
}

static void retarget_prepended_rules(
    fsa_rule_t *rules, int n, int idx_offset, int init_idx)
{
	for (fsa_rule_t *r = rules; r < rules + n; ++r) {
		if (0 == r->next)
			r->next = init_idx;
		else
			r->next += idx_offset;
	}
}

static void prepend_fsa(fsa_t *f, const fsa_t *o)
{
	assert(f != 0);

	// Ensure f's initial state has enough space for the rules from
	// o's final state.
	fsa_state_t *f_init = &f->states[f->initial];
	const fsa_state_t *o_final = &o->states[0];
	const int rule_count = f_init->count + o_final->count;
	if (f_init->capacity < rule_count) {
		do
			f_init->capacity *= 2;
		while (f_init->capacity < rule_count);
		f_init->rules
		    = realloc(f_init->rules, f_init->capacity * sizeof(fsa_rule_t));
		assert(f_init->rules);
	}

	// Copy o's final state's rules into f's intial state, then
	// retarget them.
	fsa_rule_t *start = f_init->rules + f_init->count;
	memcpy(start, o_final->rules, o_final->count * sizeof(fsa_rule_t));
	retarget_prepended_rules(
	    start, o_final->count, f->count - 1, f->initial);

	// Ensure f has enough space for the new states.
	const int count = f->count + o->count - 1;
	if (f->capacity < count) {
		do
			f->capacity *= 2;
		while (f->capacity < count);
		f->states = realloc(f->states, f->capacity * sizeof(fsa_state_t));
	}

	// Copy o's states into f, skipping index zero (the final state).
	fsa_state_t *dst = f->states + f->count;
	const fsa_state_t *src = o->states + 1;
	const int copy_count = o->count - 1;
	memcpy(dst, src, copy_count * sizeof(fsa_state_t));

	// Retarget the rules of all the newly-copied states.
	for (int i = f->count; i < count; ++i) {
		retarget_prepended_rules(
		    f->states[i].rules, f->states[i].count, f->count - 1,
		    f->initial);
	}

	// Clean up o's remaining resources.  The final state was not
	// copied to f, so that must be cleaned up along with the states
	// buffer.
	free(o->states[0].rules);
	free(o->states);

	if (0 != o->initial)
		f->initial = o->initial + f->count - 1;
	f->count = count;
}

static void construct_base(fsa_t *out)
{
	fsa_init(out);
	out->states[0].final = true;
	out->initial = fsa_add_state(out);
}

static void construct_symbol(fsa_t *out, int symbol)
{
	construct_base(out);
	fsa_add_rule(out, out->initial, 0, symbol);
}

static bool in_class(const regex_class_t *class, char c)
{
	for (int i = 0; i < class->count; ++i) {
		if (class->contents[i] == c)
			return true;
	}
	return false;
}

static void construct_class(fsa_t *out, const regex_class_t *class)
{
	construct_base(out);
	if (class->negated) {
		for (int i = 0; i < CHAR_COUNT; ++i) {
			if (!in_class(class, i))
				fsa_add_rule(out, out->initial, 0, i);
		}
	} else {
		for (int i = 0; i < class->count; ++i)
			fsa_add_rule(out, out->initial, 0, class->contents[i]);
	}
}

static void construct_wildcard(fsa_t *out)
{
	construct_base(out);
	for (int i = 0; i < CHAR_COUNT; ++i)
		fsa_add_rule(out, out->initial, 0, i);
}

static void base_quantify(fsa_t *out, int *init_out, int *final_out)
{
	fsa_t f;
	memcpy(&f, out, sizeof(fsa_t));
	construct_base(out);
	add_fsa(out, &f, init_out, final_out);
	fsa_add_rule(out, out->initial, *init_out, EPSILON);
	fsa_add_rule(out, *final_out, 0, EPSILON);
}

static void construct_star(fsa_t *out)
{
	int sub_init, sub_final;
	base_quantify(out, &sub_init, &sub_final);
	fsa_add_rule(out, sub_final, sub_init, EPSILON);
	fsa_add_rule(out, out->initial, 0, EPSILON);
}

static void construct_plus(fsa_t *out)
{
	int sub_init, sub_final;
	base_quantify(out, &sub_init, &sub_final);
	fsa_add_rule(out, sub_final, sub_init, EPSILON);
}

static void construct_qmark(fsa_t *out)
{
	int sub_init, sub_final;
	base_quantify(out, &sub_init, &sub_final);
	fsa_add_rule(out, out->initial, 0, EPSILON);
}

static void construct_term(const regex_term_t *term, fsa_t *out)
{
	switch (term->type) {
	case REGEX_TERM_EMPTY:
		construct_symbol(out, EPSILON);
		break;
	case REGEX_TERM_LITERAL:
		construct_symbol(out, term->literal);
		break;
	case REGEX_TERM_SUBEXPR:
		construct_nfa(&term->subexpr, out);
		break;
	case REGEX_TERM_CLASS:
		construct_class(out, &term->class);
		break;
	case REGEX_TERM_WILDCARD:
		construct_wildcard(out);
		break;
	}

	switch (term->quantifier) {
	case REGEX_QUANTIFIER_NONE:
		break;
	case REGEX_QUANTIFIER_STAR:
		construct_star(out);
		break;
	case REGEX_QUANTIFIER_PLUS:
		construct_plus(out);
		break;
	case REGEX_QUANTIFIER_QMARK:
		construct_qmark(out);
		break;
	}

	assert(out->states[0].final);
}

static void construct_sequence(const regex_sequence_t *seq, fsa_t *out)
{
	assert(seq->count > 0);

	fsa_t term_fsa;
	construct_term(&seq->contents[seq->count - 1], out);
	for (int i = seq->count - 2; i >= 0; --i) {
		construct_term(&seq->contents[i], &term_fsa);
		prepend_fsa(out, &term_fsa);
	}

	assert(out->states[0].final);
}

static void construct_union(fsa_t *f, const fsa_t *o)
{
	fsa_t g;
	memcpy(&g, f, sizeof(fsa_t));

	fsa_init(f);
	f->states[0].final = true;
	f->initial = fsa_add_state(f);

	int init, final;

	add_fsa(f, &g, &init, &final);
	fsa_add_rule(f, f->initial, init, EPSILON);
	fsa_add_rule(f, final, 0, EPSILON);

	add_fsa(f, o, &init, &final);
	fsa_add_rule(f, f->initial, init, EPSILON);
	fsa_add_rule(f, final, 0, EPSILON);
}

void construct_nfa(const regex_t *regex, fsa_t *out)
{
	assert(regex->count > 0);

	fsa_t sequence_fsa;
	construct_sequence(&regex->contents[0], out);
	for (int i = 1; i < regex->count; ++i) {
		construct_sequence(&regex->contents[i], &sequence_fsa);
		construct_union(out, &sequence_fsa);
	}

	assert(out->states[0].final);
}