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 construct_literal(char literal, fsa_t *out)
{
	fsa_init(out);
	const int id = fsa_add_state(out);
	fsa_add_rule(out, id, out->initial, literal);
	out->initial = id;
}

static void star_fsa(fsa_t *fsa)
{
	// If the initial state is already the final state then nothing
	// needs to be done.
	if (0 == fsa->initial)
		return;

	// Copy inital state's rules to final state.
	fsa_state_t *final = &fsa->states[0];
	const fsa_state_t *initial = &fsa->states[fsa->initial];
	if (final->capacity < final->count + initial->count) {
		do
			final->capacity *= 2;
		while (final->capacity < final->count + initial->count);
		final->rules
		    = realloc(final->rules, final->capacity * sizeof(fsa_rule_t));
		assert(final->rules);
	}
	const int copy_size = initial->count * sizeof(fsa_rule_t);
	memcpy(&final->rules[final->count], initial->rules, copy_size);
	final->count += initial->count;

	// Move states that come after initial state if there are any.
	if (fsa->count - 1 > fsa->initial) {
		const int count = fsa->count - fsa->initial - 1;
		fsa_state_t *start = &fsa->states[fsa->initial];
		memmove(start, start + 1, count * sizeof(fsa_state_t));
	}

	// Retarget all states' rules.
	for (int i = 0; i < fsa->count - 1; ++i) {
		for (int j = 0; j < fsa->states[i].count; ++j) {
			if (fsa->states[i].rules[j].next == fsa->initial)
				fsa->states[i].rules[j].next = 0;
			else if (fsa->states[i].rules[j].next > fsa->initial)
				// All states after the initial state have been moved
				// down by one position.
				--fsa->states[i].rules[j].next;
		}
	}

	--fsa->count;
	fsa->initial = 0;

	free(initial->rules);
}

static void construct_term(const regex_term_t *term, fsa_t *out)
{
	switch (term->type) {
	case REGEX_TERM_EMPTY:
		fsa_init(out);
		break;
	case REGEX_TERM_LITERAL:
		construct_literal(term->literal, out);
		break;
	case REGEX_TERM_SUBEXPR:
		construct(&term->subexpr, out);
		break;
	case REGEX_TERM_WILDCARD:
	case REGEX_TERM_CLASS:
		assert(false);
		break;
	}

	switch (term->quantifier) {
	case REGEX_QUANTIFIER_NONE:
		break;
	case REGEX_QUANTIFIER_STAR:
		star_fsa(out);
		break;
	case REGEX_QUANTIFIER_PLUS:
	case REGEX_QUANTIFIER_QMARK:
		assert(false);
		break;
	}

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

static void concat_fsas(fsa_t *base, const fsa_t *other)
{
	// TODO: Handle the other's final state having transition rules.
	assert(0 == other->states[0].count);

	// Copy states other than the final state (index zero) to base.
	const int new_count = base->count + other->count - 1;
	if (base->capacity < new_count) {
		do
			base->capacity *= 2;
		while (base->capacity < new_count);
		base->states = realloc(base->states, base->capacity);
		assert(base->states);
	}
	const int copy_size = (other->count - 1) * sizeof(fsa_state_t);
	memcpy(&base->states[base->count], &other->states[1], copy_size);

	// Retarget new states' rules.
	for (int i = base->count; i < new_count; ++i) {
		fsa_state_t *state = &base->states[i];
		for (int j = 0; j < state->count; ++j) {
			if (0 == state->rules[j].next)
				state->rules[j].next = base->initial;
			else
				// States' indices have increased by one less than the
				// base count, as the final state came before them and
				// was not copied.
				state->rules[j].next += base->count - 1;
		}
	}

	base->initial = other->initial + base->count - 1;
	base->count = new_count;

	free(other->states[0].rules);
	free(other->states);

	assert(base->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);
		concat_fsas(out, &term_fsa);
	}

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

static void retarget_merged_rules(
    fsa_rule_t *rules, int rules_count, int initial, int base_initial,
    int base_count)
{
	for (int i = 0; i < rules_count; ++i) {
		if (0 == rules[i].next)
			continue;

		// If the state came before the initial state it should be
		// offset by one less than base_count, because the final state
		// (index zero) came before it and was not copied into the
		// base.
		const int before_offset = base_count - 1;

		// If it came after the initial state it must be offset by two
		// less than base_count because both the final state and the
		// initial state came before it and were not copied -- unless
		// the initial state is the same state as the final state, in
		// which case the offset is still only one less than
		// base_count.
		const int after_offset = base_count - (0 != initial ? 2 : 1);

		if (rules[i].next < initial)
			rules[i].next += before_offset;
		else if (rules[i].next > initial)
			rules[i].next += after_offset;
		else if (rules[i].next == initial)
			rules[i].next = base_initial;
	}
}

static void merge_fsas(fsa_t *base, const fsa_t *other)
{
	// Copy rules from the other's initial state into the base's
	// initial state.
	fsa_state_t *initial = &base->states[base->initial];
	const fsa_state_t *other_initial = &other->states[other->initial];
	const int new_rule_count = initial->count + other_initial->count;
	if (initial->capacity < new_rule_count) {
		do
			initial->capacity *= 2;
		while (initial->capacity < new_rule_count);
		initial->rules = realloc(
		    initial->rules, initial->capacity * sizeof(fsa_rule_t));
		assert(initial->rules);
	}
	memcpy(
	    &initial->rules[initial->count], other_initial->rules,
	    other_initial->count * sizeof(fsa_rule_t));

	// Retarget the copied rules.
	retarget_merged_rules(
	    &initial->rules[initial->count], other_initial->count,
	    other->initial, base->initial, base->count);

	// Copy other states, skipping the initial state.
	const int skipped_states = other->initial != 0 ? 2 : 1;
	const int new_count = base->count + other->count - skipped_states;
	if (base->capacity < new_count) {
		do
			base->capacity *= 2;
		while (base->capacity < new_count);
		base->states
		    = realloc(base->states, base->capacity * sizeof(fsa_state_t));
		assert(base->states);
	}
	int offset = base->count;
	if (1 < other->initial) {
		const int copy_count = other->initial - 1;
		const int copy_size = copy_count * sizeof(fsa_state_t);
		memcpy(&base->states[offset], &other->states[1], copy_size);
		offset += copy_count;
	}
	if (other->initial < other->count - 1) {
		const int copy_count = other->count - other->initial - 1;
		const int copy_size = copy_count * sizeof(fsa_state_t);
		memcpy(
		    &base->states[offset], &other->states[other->initial],
		    copy_size);
	}

	// Retarget the copied states' rules.
	for (int i = base->count; i < new_count; ++i) {
		retarget_merged_rules(
		    base->states[i].rules, base->states[i].count, other->initial,
		    base->initial, base->count);
	}

	initial->count = new_rule_count;
	base->count = new_count;

	free(other->states[0].rules);
	if (other->initial != 0)
		free(other->states[other->initial].rules);
	free(other->states);

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

void construct(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);
		merge_fsas(out, &sequence_fsa);
	}

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