[paraslash.git] / mp.c

/*
 * Copyright (C) 2017 Andre Noll <maan@tuebingen.mpg.de>
 *
 * Licensed under the GPL v2. For licencing details see COPYING.
 */

/**
 * \file mp.c Mood parser helper functions.
 *
 * This file contains the public and the private API of the flex/bison based
 * mood parser.
 *
 * The public API (at the bottom of the file) allows to parse the same mood
 * definition many times in an efficient manner.
 *
 * The first function to all is \ref mp_init(), which analyzes the given mood
 * definition syntactically. It returns the abstract syntax tree of the mood
 * definition and pre-compiles all regular expression patterns to make later
 * pattern matching efficient.
 *
 * Semantic analysis is performed in \ref mp_eval_row(). This function is
 * called from \ref mood.c once for each file in the audio file table. It
 * utilizes the abstract syntax tree and the pre-compiled regular expressions
 * to determine the set of admissible audio files.
 *
 * If the mood is no longer needed, \ref mp_shutdown() should be called to free
 * the resources.
 *
 * The internal API is described in \ref mp.h.
 */

#include "para.h"

#include <regex.h>
#include <fnmatch.h>
#include <osl.h>
#include <lopsub.h>

#include "string.h"
#include "error.h"
#include "afh.h"
#include "afs.h"
#include "mp.h"
#include "mp.bison.h"

struct mp_context {
        /* global context */
        char *errmsg;
        struct mp_ast_node *ast;
        /* per audio file context */
        const struct osl_row *aft_row;
        char *path;
        bool have_afsi;
        struct afs_info afsi;
        bool have_afhi;
        struct afh_info afhi;
};

/**
 * Parse a (generalized) string literal.
 *
 * \param src The string to parse.
 * \param quote_chars Opening and closing quote characters.
 * \param result The corresponding C string is returned here.
 *
 * This function turns a generalized C99 string literal like "xyz\n" into a C
 * string (containing the three characters 'x', 'y' and 'z', followed by a
 * newline character and the terminating zero byte). The function allows to
 * specify different quote characters so that, for example, regular expression
 * patterns enclosed in '/' can be parsed as well. To parse a proper string
 * literal, one has to pass two double quotes as the second argument.
 *
 * The function strips off the opening and leading quote characters, replaces
 * double backslashes by single backslashes and handles the usual escapes like
 * \n and \".
 *
 * The caller must make sure that the input is well-formed. The function simply
 * aborts if the input is not a valid C99 string literal (modulo the quote
 * characters).
 *
 * \return Offset of the first character after the closing quote. For proper
 * string literals this will be the terminating zero byte of the input string,
 * for regular expression patterns it is the beginning of the flags which
 * modify the matching behaviour.
 *
 * \sa \ref mp_parse_regex_pattern(), \ref mp_parse_wildcard_pattern().
 */
unsigned parse_quoted_string(const char *src, const char quote_chars[2],
                char **result)
{
        size_t n, len = strlen(src);
        char *dst, *p;
        bool backslash;

        assert(len >= 2);
        assert(src[0] == quote_chars[0]);
        p = dst = para_malloc(len - 1);
        backslash = false;
        for (n = 1;; n++) {
                char c;
                assert(n < len);
                c = src[n];
                if (!backslash) {
                        if (c == '\\') {
                                backslash = true;
                                continue;
                        }
                        if (c == quote_chars[1])
                                break;
                        *p++ = c;
                        continue;
                }
                if (c == quote_chars[1])
                        *p++ = quote_chars[1];
                else switch (c) {
                        case '\\': *p++ = '\\'; break;
                        case 'a': *p++ = '\a'; break;
                        case 'b': *p++ = '\b'; break;
                        case 'f': *p++ = '\f'; break;
                        case 'n': *p++ = '\n'; break;
                        case 'r': *p++ = '\r'; break;
                        case 't': *p++ = '\t'; break;
                        case 'v': *p++ = '\v'; break;
                        default: assert(false);
                }
                backslash = false;
        }
        assert(src[n] == quote_chars[1]);
        *p = '\0';
        *result = dst;
        return n + 1;
}

/**
 * Parse and compile an extended regular expression pattern, including flags.
 *
 * \param src The pattern to parse.
 * \param result C-string and flags are returned here.
 *
 * A regex pattern is identical to a C99 string literal except (a) it is
 * enclosed in '/' characters rather than double quotes, (b) double quote
 * characters which are part of the pattern do not need to be quoted with
 * backslashes, but slashes must be quoted in this way, and (c) the closing
 * slash may be followed by one or more flag characters which modify the
 * matching behaviour.
 *
 * The only flags which are currently supported are 'i' to ignore case in match
 * (REG_ICASE) and 'n' to change the handling of newline characters
 * (REG_NEWLINE).
 *
 * \return Standard. This function calls \ref parse_quoted_string(), hence it
 * aborts if the input string is malformed. However, errors from \ref
 * para_regcomp are returned without aborting the process. The rationale behind
 * this difference is that passing a malformed string must be considered an
 * implementation bug because malformed strings should be rejected earlier by
 * the lexer.
 *
 * \sa \ref mp_parse_wildcard_pattern(), \ref parse_quoted_string(),
 * \ref para_regcomp(), regex(3).
 */
int mp_parse_regex_pattern(const char *src, struct mp_re_pattern *result)
{
        int ret;
        char *pat;
        unsigned n = parse_quoted_string(src, "//", &pat);

        result->flags = 0;
        for (; src[n]; n++) {
                switch (src[n]) {
                case 'i': result->flags |= REG_ICASE; break;
                case 'n': result->flags |= REG_NEWLINE; break;
                default: assert(false);
                }
        }
        ret = para_regcomp(&result->preg, pat, result->flags);
        free(pat);
        return ret;
}

/**
 * Parse a wildcard pattern, including flags.
 *
 * \param src The pattern to parse.
 * \param result C-string and flags are returned here.
 *
 * This function parses a shell wildcard pattern. It is similar to \ref
 * mp_parse_regex_pattern(), so the remarks mentioned there apply to this
 * function as well.
 *
 * Wildcard patterns differ from regular expression patterns in that (a) they
 * must be enclosed in '|' characters, (b) they support different flags for
 * modifying matching behaviour, and (c) there is no cache for them.
 *
 * The following flags, whose meaning is explained in fnmatch(3), are currently
 * supported: 'n' (FNM_NOESCAPE), 'p' (FNM_PATHNAME), 'P' (FNM_PERIOD), 'l'
 * (FNM_LEADING_DIR), 'i' (FNM_CASEFOLD), 'e' (FNM_EXTMATCH). The last flag is
 * a GNU extension. It is silently ignored on non GNU systems.
 *
 * \sa \ref parse_quoted_string(), \ref mp_parse_regex_pattern(), fnmatch(3).
 */
void mp_parse_wildcard_pattern(const char *src, struct mp_wc_pattern *result)
{
        unsigned n = parse_quoted_string(src, "||", &result->pat);

        result->flags = 0;
        for (; src[n]; n++) {
                switch (src[n]) {
                case 'n': result->flags |= FNM_NOESCAPE; break;
                case 'p': result->flags |= FNM_PATHNAME; break;
                case 'P': result->flags |= FNM_PERIOD; break;
                /* not POSIX, but both FreeBSD and NetBSD have it */
                case 'l': result->flags |= FNM_LEADING_DIR; break;
                case 'i': result->flags |= FNM_CASEFOLD; break;
                /* GNU only */
#ifdef HAVE_FNM_EXTMATCH
                case 'e': result->flags |= FNM_EXTMATCH; break;
#else /* silently ignore extglob flag */
                case 'e': break;
#endif
                default: assert(false);
                }
        }
}

/**
 * Set the error bit in the parser context and log a message.
 *
 * \param line The number of the input line which caused the error.
 * \param ctx Contains the error bit.
 * \param fmt Usual format string.
 *
 * This is called if the lexer or the parser detect an error in the mood
 * definition. Only the first error is logged (with a severity of "warn").
 */
__printf_3_4 void mp_parse_error(int line, struct mp_context *ctx,
                const char *fmt, ...)
{
        va_list ap;
        char *tmp;

        if (ctx->errmsg) /* we already printed an error message */
                return;
        va_start(ap, fmt);
        xvasprintf(&tmp, fmt, ap);
        va_end(ap);
        xasprintf(&ctx->errmsg, "line %d: %s", line, tmp);
        free(tmp);
        PARA_WARNING_LOG("%s\n", ctx->errmsg);
}

static int get_afsi(struct mp_context *ctx)
{
        int ret;

        if (ctx->have_afsi)
                return 0;
        ret = get_afsi_of_row(ctx->aft_row, &ctx->afsi);
        if (ret < 0)
                return ret;
        ctx->have_afsi = true;
        return 1;
}

static int get_afhi(struct mp_context *ctx)
{
        int ret;

        if (ctx->have_afhi)
                return 0;
        ret = get_afhi_of_row(ctx->aft_row, &ctx->afhi);
        if (ret < 0)
                return ret;
        ctx->have_afhi = true;
        return 1;
}

/**
 * Return the full path to the audio file.
 *
 * \param ctx Contains a reference to the row of the audio file table which
 * corresponds to the current audio file. The path of the audio file, the
 * afs_info and the afh_info structures (which contain the tag information) can
 * be retrieved through this reference.
 *
 * \return A reference to the path. Must not be freed by the caller.
 *
 * \sa \ref get_audio_file_path_of_row().
 */
char *mp_path(struct mp_context *ctx)
{
        if (!ctx->path)
                get_audio_file_path_of_row(ctx->aft_row, &ctx->path);
        return ctx->path;
}

/**
 * Check whether the given attribute is set for the current audio file.
 *
 * \param attr The string to look up in the attribute table.
 * \param ctx See \ref mp_path().
 *
 * First, determine the bit number which corresponds to the attribute, then
 * check if this bit is set in the ->attributes field of the afs_info structure
 * of the audio file.
 *
 * \return True if the attribute is set, false if it is not. On errors, for
 * example if the given string is no attribute, the function returns false.
 *
 * \sa \ref get_attribute_bitnum_by_name().
 */
bool mp_is_set(const char *attr, struct mp_context *ctx)
{
        int ret;
        unsigned char bitnum;
        const uint64_t one = 1;

        ret = get_attribute_bitnum_by_name(attr, &bitnum);
        if (ret < 0) /* treat invalid attributes as not set */
                return false;
        ret = get_afsi(ctx);
        if (ret < 0)
                return false;
        return (one << bitnum) & ctx->afsi.attributes;
}

/**
 * Count the number of attributes set.
 *
 * \param ctx See \ref mp_path().
 *
 * \return The number of bits which are set in the ->attributes field of the
 * afs_info structure of the current audio file.
 */
int64_t mp_num_attributes_set(struct mp_context *ctx)
{
        const uint64_t m = ~(uint64_t)0;
        int ret;
        uint64_t v;

        ret = get_afsi(ctx);
        if (ret < 0)
                return 0;

        v = ctx->afsi.attributes;
        /* taken from https://graphics.stanford.edu/~seander/bithacks.html */
        v = v - ((v >> 1) & m / 3);
        v = (v & m / 15 * 3) + ((v >> 2) & m / 15 * 3);
        v = (v + (v >> 4)) & m / 255 * 15;
        v = (v * (m / 255)) >> 56;
        assert(v <= 64);
        return v;
}

/**
 * Define a function which returns a field of the afs_info structure.
 *
 * \param _name The name of the field.
 *
 * The defined function casts the value to int64_t. On errors, zero is returned.
 */
#define MP_AFSI(_name) \
        int64_t mp_ ## _name(struct mp_context *ctx) \
        { \
                int ret = get_afsi(ctx); \
                if (ret < 0) \
                        return 0; \
                return ctx->afsi._name; \
        }
/** \cond MP_AFSI */
MP_AFSI(num_played)
MP_AFSI(image_id)
MP_AFSI(lyrics_id)
/** \endcond */

/**
 * Define a function which returns a field of the afh_info structure.
 *
 * \param _name The name of the field.
 *
 * The defined function casts the value to int64_t. On errors, zero is returned.
 */
#define MP_AFHI(_name) \
        int64_t mp_ ## _name(struct mp_context *ctx) \
        { \
                int ret = get_afhi(ctx); \
                if (ret < 0) \
                        return 0; \
                return ctx->afhi._name; \
        }
/** \cond MP_AFHI */
MP_AFHI(bitrate)
MP_AFHI(frequency)
MP_AFHI(channels)
/** \endcond */

/**
 * Define a function which extracts and returns the value of a meta tag.
 *
 * \param _name The name of the tag (artist, title, ...).
 *
 * The function will return a pointer to memory owned by the audio file
 * selector. On errors, or if the current audio file has no tag of the given
 * name, the function returns the empty string. The caller must not attempt to
 * free the returned string.
 */
#define MP_TAG(_name) \
        char *mp_ ## _name (struct mp_context *ctx) \
        { \
                int ret = get_afhi(ctx); \
                if (ret < 0) \
                        return ""; \
                return ctx->afhi.tags._name; \
        }
/** \cond MP_TAG */
MP_TAG(artist)
MP_TAG(title)
MP_TAG(album)
MP_TAG(comment)
/** \endcond */

/**
 * Parse and return the value of the year tag.
 *
 * \param ctx See \ref mp_path().
 *
 * \return If the year tag is not present, can not be parsed, or its value is
 * less than zero, the function returns 0. If the value is less than 100, we
 * add 1900.
 */
int64_t mp_year(struct mp_context *ctx)
{
        int64_t year;
        int ret = get_afhi(ctx);

        if (ret < 0)
                return 0;
        assert(ctx->afhi.tags.year);
        ret = para_atoi64(ctx->afhi.tags.year, &year);
        if (ret < 0)
                return 0;
        if (year < 0)
                return 0;
        if (year < 100)
                year += 1900;
        return year;
}

/*
 * Ideally, these functions should be declared in a header file which is
 * created by flex with the --header-file option. However, for flex-2.6.x
 * (2017) this option is borken: if --reentrant is also given, the generated
 * header file contains syntax errors. As a workaround we declare the functions
 * here.
 */
/** \cond flex_workaround */
int mp_yylex_init(mp_yyscan_t *yyscanner);
struct yy_buffer_state *mp_yy_scan_bytes(const char *buf, int len,
        mp_yyscan_t yyscanner);
void mp_yy_delete_buffer(struct yy_buffer_state *bs, mp_yyscan_t yyscanner);
int mp_yylex_destroy(mp_yyscan_t yyscanner);
void mp_yyset_lineno(int lineno, mp_yyscan_t scanner);
/** \endcond */

/* Public API */

/**
 * Initialize the mood parser.
 *
 * This allocates and sets up the internal structures of the mood parser
 * and creates an abstract syntax tree from the given mood definition.
 * It must be called before \ref mp_eval_row() can be called.
 *
 * The context pointer returned by this function may be passed to \ref
 * mp_eval_row() to determine whether an audio file is admissible.
 *
 * \param definition A reference to the mood definition.
 * \param nbytes The size of the mood definition.
 * \param result Opaque context pointer is returned here.
 * \param errmsg Optional error message is returned here.
 *
 * It's OK to pass a NULL pointer or a zero sized buffer as the mood
 * definition. This corresponds to the "dummy" mood for which all audio files
 * are admissible.
 *
 * The error message pointer may also be NULL in which case no error message
 * is returned. Otherwise, the caller must free the returned string.
 *
 * \return Standard. On success *errmsg is set to NULL.
 */
int mp_init(const char *definition, int nbytes, struct mp_context **result,
                 char **errmsg)
{
        int ret;
        mp_yyscan_t scanner;
        struct mp_context *ctx;
        struct yy_buffer_state *buffer_state;

        if (!definition || nbytes == 0) { /* dummy mood */
                if (errmsg)
                        *errmsg = NULL;
                *result = NULL;
                return 0;
        }
        ctx = para_calloc(sizeof(*ctx));
        ctx->errmsg = NULL;
        ctx->ast = NULL;

        ret = mp_yylex_init(&scanner);
        assert(ret == 0);
        buffer_state = mp_yy_scan_bytes(definition, nbytes, scanner);
        mp_yyset_lineno(1, scanner);
        PARA_NOTICE_LOG("creating abstract syntax tree\n");
        ret = mp_yyparse(ctx, &ctx->ast, scanner);
        mp_yy_delete_buffer(buffer_state, scanner);
        mp_yylex_destroy(scanner);
        if (ctx->errmsg) { /* parse error */
                if (errmsg)
                        *errmsg = ctx->errmsg;
                else
                        free(ctx->errmsg);
                free(ctx);
                return -E_MOOD_PARSE;
        }
        if (errmsg)
                *errmsg = NULL;
        *result = ctx;
        return 1;
}

/**
 * Determine whether the given audio file is admissible.
 *
 * \param aft_row The audio file to check for admissibility.
 * \param ctx As returned from \ref mp_init().
 *
 * \return Whether the audio file is admissible.
 *
 * If the mood parser was set up without an input buffer (dummy mood), this
 * function returns true (without looking at the audio file metadata) to
 * indicate that the given audio file should be considered admissible.
 *
 * \sa \ref change_current_mood(), \ref mp_eval_ast().
 */
bool mp_eval_row(const struct osl_row *aft_row, struct mp_context *ctx)
{
        if (!ctx) /* dummy mood */
                return true;
        assert(aft_row);
        ctx->aft_row = aft_row;
        ctx->have_afsi = false;
        ctx->have_afhi = false;
        ctx->path = NULL;
        return mp_eval_ast(ctx->ast, ctx);
}

/**
 * Deallocate the resources of a mood parser.
 *
 * This function frees the abstract syntax tree which was created by \ref
 * mp_init().
 *
 * \param ctx As returned from \ref mp_init().
 *
 * It's OK to pass a NULL pointer, in which case the function does nothing.
 */
void mp_shutdown(struct mp_context *ctx)
{
        if (!ctx)
                return;
        mp_free_ast(ctx->ast);
        free(ctx);
}