X-Git-Url: https://sigrok.org/gitweb/?p=libsigrok.git;a=blobdiff_plain;f=src%2Finput%2Fcsv.c;h=66cf4bb3c2445af3ad9c8c3c1be36e2d658b03d4;hp=ba50e2ca8b5286c5ec2536be4ca9df8de398251f;hb=cd11e33c5de58edb0340fd9f25115aa4509f4767;hpb=43e1e23a115e00d22d814da6a00e16015e726213 diff --git a/src/input/csv.c b/src/input/csv.c index ba50e2ca..66cf4bb3 100644 --- a/src/input/csv.c +++ b/src/input/csv.c @@ -20,9 +20,11 @@ #include "config.h" +#include #include #include #include +#include #include #include "libsigrok-internal.h" @@ -125,27 +127,15 @@ * and equidistant rows, there is no fancy support for textual unit * suffixes nor gaps in the stream of samples nor other non-linearity, * just '-' ignore the column if the format is not supported). + * + * IMPORTANT! Make sure the .format_match() logic matches the default + * values for the input module's options. Ideally the format match test + * shall pass for all input data that is supported by default. */ /* * TODO * - * - Extend support for analog input data. - * - Determine why analog samples of 'double' data type get scrambled - * in sigrok-cli screen output. Is analog.encoding->unitsize not - * handled properly? A sigrok-cli or libsigrok (src/output) issue? - * - Reconsider the channel creation after format processing. Current - * logic may "bleed" channel names into the analog group when logic - * channels' columns follow analog columns (seen with "-,2a,x8"). - * Trying to sort it out, a naive change used to map logic channels' - * data to incorrect bitmap positions. The whole channel numbering - * needs reconsideration. Probably it's easiest to first create _all_ - * logic channels so that they have adjacent numbers starting at 0 - * (addressing logic bits), then all analog channels (again adjacent) - * to simplify the calculation of their index in the sample set as - * well as their sdi channel index from the "analog column index". - * - Optionally get sample rate from timestamp column. Just best-effort - * approach, not necessarily reliable. Users can always specify rates. * - Add a test suite for input modules in general, and CSV in specific? * Becomes more important with the multitude of options and their * interaction. Could cover edge cases (BOM presence, line termination @@ -153,7 +143,7 @@ * samplerates, etc). */ -typedef float csv_analog_t; /* 'double' currently is flawed. */ +typedef double csv_analog_t; /* Single column formats. */ enum single_col_format { @@ -208,8 +198,8 @@ struct column_details { enum single_col_format text_format; size_t channel_offset; size_t channel_count; - size_t channel_index; int analog_digits; + GString **channel_names; }; struct context { @@ -217,6 +207,7 @@ struct context { /* Current samplerate, optionally determined from input data. */ uint64_t samplerate; + uint64_t calc_samplerate; double prev_timestamp; gboolean samplerate_sent; @@ -256,14 +247,15 @@ struct context { csv_analog_t *analog_datafeed_buffer; /**!< Queue for analog datafeed. */ size_t analog_datafeed_buf_size; size_t analog_datafeed_buf_fill; - GSList **analog_datafeed_channels; int *analog_datafeed_digits; + GSList **analog_datafeed_channels; /* Current line number. */ size_t line_number; /* List of previously created sigrok channels. */ GSList *prev_sr_channels; + GSList **prev_df_channels; }; /* @@ -282,19 +274,13 @@ struct context { static int flush_samplerate(const struct sr_input *in) { struct context *inc; - struct sr_datafeed_packet packet; - struct sr_datafeed_meta meta; - struct sr_config *src; inc = in->priv; - if (inc->samplerate && !inc->samplerate_sent) { - packet.type = SR_DF_META; - packet.payload = &meta; - src = sr_config_new(SR_CONF_SAMPLERATE, g_variant_new_uint64(inc->samplerate)); - meta.config = g_slist_append(NULL, src); - sr_session_send(in->sdi, &packet); - g_slist_free(meta.config); - sr_config_free(src); + if (!inc->calc_samplerate && inc->samplerate) + inc->calc_samplerate = inc->samplerate; + if (inc->calc_samplerate && !inc->samplerate_sent) { + (void)sr_session_send_meta(in->sdi, SR_CONF_SAMPLERATE, + g_variant_new_uint64(inc->calc_samplerate)); inc->samplerate_sent = TRUE; } @@ -353,6 +339,7 @@ static int flush_logic_samples(const struct sr_input *in) return rc; inc->datafeed_buf_fill = 0; + return SR_OK; } @@ -371,6 +358,7 @@ static int queue_logic_samples(const struct sr_input *in) if (rc != SR_OK) return rc; } + return SR_OK; } @@ -446,6 +434,7 @@ static int flush_analog_samples(const struct sr_input *in) } inc->analog_datafeed_buf_fill = 0; + return SR_OK; } @@ -464,6 +453,7 @@ static int queue_analog_samples(const struct sr_input *in) if (rc != SR_OK) return rc; } + return SR_OK; } @@ -563,12 +553,13 @@ static int make_column_details_from_format(const struct sr_input *in, char *column; const char *caption; int channel_type, channel_sdi_nr; + void *channel; int ret; inc = in->priv; inc->column_seen_count = g_strv_length(column_texts); - /* Split the input spec, count involved columns and bits. */ + /* Split the input spec, count involved columns and channels. */ formats = g_strsplit(column_format, ",", 0); if (!formats) { sr_err("Cannot parse columns format %s (comma split).", column_format); @@ -610,7 +601,7 @@ static int make_column_details_from_format(const struct sr_input *in, sr_dbg("Column format %s -> %zu columns, %zu logic, %zu analog channels.", column_format, column_count, logic_count, analog_count); - /* Allocate and fill in "column processing" details. Create channels. */ + /* Allocate and fill in "column processing" details. */ inc->column_want_count = column_count; if (inc->column_seen_count < inc->column_want_count) { sr_err("Insufficient input text width for desired data amount, got %zu but want %zu columns.", @@ -618,6 +609,10 @@ static int make_column_details_from_format(const struct sr_input *in, g_strfreev(formats); return SR_ERR_ARG; } + inc->logic_channels = logic_count; + inc->analog_channels = analog_count; + inc->analog_datafeed_digits = g_malloc0(inc->analog_channels * sizeof(inc->analog_datafeed_digits[0])); + inc->analog_datafeed_channels = g_malloc0(inc->analog_channels * sizeof(inc->analog_datafeed_channels[0])); inc->column_details = g_malloc0_n(column_count, sizeof(inc->column_details[0])); column_idx = channel_idx = analog_idx = 0; channel_name = g_string_sized_new(64); @@ -669,12 +664,12 @@ static int make_column_details_from_format(const struct sr_input *in, if (!caption || !*caption) caption = NULL; /* - * TODO Need we first create _all_ logic channels, - * before creating analog channels? Just store the - * parameters here (index, type, name) and have the - * creation sequence done outside of the format - * spec parse loop. + * Collect channel creation details here, but defer + * actual creation of the channels such that all + * logic channels can get created first and analog + * channels only get created afterwards. */ + detail->channel_names = g_malloc0(detail->channel_count * sizeof(detail->channel_names[0])); for (create_idx = 0; create_idx < detail->channel_count; create_idx++) { if (caption && detail->channel_count == 1) { g_string_assign(channel_name, caption); @@ -685,26 +680,38 @@ static int make_column_details_from_format(const struct sr_input *in, g_string_printf(channel_name, "%zu", detail->channel_offset + create_idx); } - if (format_is_analog(detail->text_format)) { - channel_sdi_nr = logic_count + detail->channel_offset + create_idx; - channel_type = SR_CHANNEL_ANALOG; - detail->channel_index = g_slist_length(in->sdi->channels); - } else if (format_is_logic(detail->text_format)) { - channel_sdi_nr = detail->channel_offset + create_idx; - channel_type = SR_CHANNEL_LOGIC; - } else { - continue; - } - sr_channel_new(in->sdi, channel_sdi_nr, - channel_type, TRUE, channel_name->str); + detail->channel_names[create_idx] = g_string_new_len(channel_name->str, channel_name->len); } } } - inc->logic_channels = channel_idx; - inc->analog_channels = analog_idx; g_string_free(channel_name, TRUE); g_strfreev(formats); + /* Create channels in strict logic to analog order. */ + channel_type = SR_CHANNEL_LOGIC; + for (column_idx = 0; column_idx < inc->column_want_count; column_idx++) { + detail = &inc->column_details[column_idx]; + if (!format_is_logic(detail->text_format)) + continue; + for (create_idx = 0; create_idx < detail->channel_count; create_idx++) { + caption = detail->channel_names[create_idx]->str; + channel_sdi_nr = g_slist_length(in->sdi->channels); + sr_channel_new(in->sdi, channel_sdi_nr, channel_type, TRUE, caption); + } + } + channel_type = SR_CHANNEL_ANALOG; + for (column_idx = 0; column_idx < inc->column_want_count; column_idx++) { + detail = &inc->column_details[column_idx]; + if (!format_is_analog(detail->text_format)) + continue; + caption = detail->channel_names[0]->str; + channel_sdi_nr = g_slist_length(in->sdi->channels); + channel = sr_channel_new(in->sdi, channel_sdi_nr, channel_type, TRUE, caption); + channel_idx = channel_sdi_nr - inc->logic_channels; + inc->analog_datafeed_digits[channel_idx] = detail->analog_digits; + inc->analog_datafeed_channels[channel_idx] = g_slist_append(NULL, channel); + } + return SR_OK; } @@ -714,6 +721,7 @@ static const struct column_details *lookup_column_details(struct context *inc, s return NULL; if (!nr || nr > inc->column_want_count) return NULL; + return &inc->column_details[nr - 1]; } @@ -737,7 +745,7 @@ static void strip_comment(char *buf, const GString *prefix) } /** - * @brief Splits a text line into a set of columns. + * Splits a text line into a set of columns. * * @param[in] buf The input text line to split. * @param[in] inc The input module's context. @@ -748,11 +756,24 @@ static void strip_comment(char *buf, const GString *prefix) */ static char **split_line(char *buf, struct context *inc) { - return g_strsplit(buf, inc->delimiter->str, 0); + char **fields, *f; + size_t l; + + fields = g_strsplit(buf, inc->delimiter->str, 0); + if (!fields) + return NULL; + + l = g_strv_length(fields); + while (l--) { + f = fields[l]; + g_strchomp(f); + } + + return fields; } /** - * @brief Parse a multi-bit field into several logic channels. + * Parse a multi-bit field into several logic channels. * * @param[in] column The input text, a run of bin/hex/oct digits. * @param[in] inc The input module's context. @@ -862,7 +883,7 @@ static int parse_logic(const char *column, struct context *inc, } /** - * @brief Parse a floating point text into an analog value. + * Parse a floating point text into an analog value. * * @param[in] column The input text, a floating point number. * @param[in] inc The input module's context. @@ -911,7 +932,7 @@ static int parse_analog(const char *column, struct context *inc, } /** - * @brief Parse a timestamp text, auto-determine samplerate. + * Parse a timestamp text, auto-determine samplerate. * * @param[in] column The input text, a floating point number. * @param[in] inc The input module's context. @@ -962,13 +983,13 @@ static int parse_timestamp(const char *column, struct context *inc, * reduced rounding errors which result in odd rates. * - Support other formats ("2 ms" or similar)? */ - if (inc->samplerate) + if (inc->calc_samplerate) return SR_OK; ret = sr_atod_ascii(column, &ts); if (ret != SR_OK) ts = 0.0; if (!ts) { - sr_warn("Cannot convert timestamp text %s in line %zu (or zero value).", + sr_info("Cannot convert timestamp text %s in line %zu (or zero value).", column, inc->line_number); inc->prev_timestamp = 0.0; return SR_OK; @@ -993,14 +1014,14 @@ static int parse_timestamp(const char *column, struct context *inc, rate += 0.5; rate = (uint64_t)rate; sr_dbg("Rate from timestamp %g in line %zu.", rate, inc->line_number); - inc->samplerate = rate; + inc->calc_samplerate = rate; inc->prev_timestamp = 0.0; return SR_OK; } /** - * @brief Parse routine which ignores the input text. + * Parse routine which ignores the input text. * * This routine exists to unify dispatch code paths, mapping input file * columns' data types to their respective parse routines. @@ -1011,6 +1032,7 @@ static int parse_ignore(const char *column, struct context *inc, (void)column; (void)inc; (void)details; + return SR_OK; } @@ -1026,6 +1048,126 @@ static const col_parse_cb col_parse_funcs[] = { [FORMAT_TIME] = parse_timestamp, }; +/* + * BEWARE! Implementor's notes. Sync with feature set and default option + * values required during maintenance of the input module implementation. + * + * When applications invoke .format_match() routines, trying automatic + * determination of an input file's format handler, then no options are + * in effect. Because specifying options requires selection of an input + * module to pass the options to, which obsoletes the format-match check. + * + * Which means that we only need to deal with the default format here, + * which happens to be the simple multi-column format without header + * lines or leading garbage. Which means that the check can be rather + * strict, resulting in high levels of confidence upon match, never + * "accidently" winning for unreadable or unsupported-by-default formats. + * + * This .format_match() logic only needs to become more involved when + * default option values change, or when automatic detection of column + * data types improves. Then the supported-by-default types of input + * data must be considered acceptable here in the format-match check + * as well. + * + * Notice that the format check cannot re-use regular processing logic + * when their implementation assumes proper input data and wll generate + * diagnostics for unexpected input data. Failure to match the format is + * non-fatal here, mismatch must remain silent. It's up to applications + * how large a chunk of data gets passed here (start of the file's + * content). But inspection of the first few hundred bytes will usually + * be GoodEnough(TM) for the format-match purpose. Notice that filenames + * need not necessarily be available to the format-match routine. + * + * This implementation errs on the safe side. Users can always select + * the CSV input module when automatic format detection fails. + */ +static int format_match(GHashTable *metadata, unsigned int *confidence) +{ + const int match_confidence = 100; + const char *default_extension = ".csv"; + const char *line_termination = "\n"; + const char *comment_leader = ";"; + const char *column_separator = ","; + const char *binary_charset = "01"; + + const char *fn; + GString *buf; + size_t fn_len; + GString *tmpbuf; + gboolean status; + size_t line_idx, col_idx; + char *rdptr, **lines, *line; + char **cols, *col; + + /* Get the application provided input data properties. */ + fn = g_hash_table_lookup(metadata, GINT_TO_POINTER(SR_INPUT_META_FILENAME)); + buf = g_hash_table_lookup(metadata, GINT_TO_POINTER(SR_INPUT_META_HEADER)); + + /* Filenames are a strong hint. Use then when available. */ + if (fn && *fn && (fn_len = strlen(fn)) >= strlen(default_extension)) { + if (strcasecmp(&fn[fn_len - strlen(default_extension)], default_extension) == 0) { + *confidence = 10; + return SR_OK; + } + } + + /* + * Check file content for compatibility with the input module's + * default format. Which translates to: + * - Must be at least one text line worth of input data. Ignore + * incomplete lines at the end of the available buffer. + * - Must be LF terminated text lines, optional CR-LF sequence. + * (Drop CR-only for simplicity since that's rare and users + * can override the automatic detection.) + * - Strip comments and skip empty lines. + * - Data lines must be binary input (potentially multiple bits + * per column which then get ignored). Presence of comma is + * optional but then must be followed by another data column. + * - No other content is acceptable, there neither are ignored + * columns nor analog data nor timestamps in the default layout. + * (See the above "sync format match with default options" + * comment though during maintenance!) + * Run the check on a copy to not affect the caller's buffer. + */ + if (!buf || !buf->len || !buf->str || !*buf->str) + return SR_ERR; + rdptr = g_strstr_len(buf->str, buf->len, line_termination); + if (!rdptr) + return SR_ERR; + tmpbuf = g_string_new_len(buf->str, rdptr + 1 - buf->str); + tmpbuf->str[tmpbuf->len - 1] = '\0'; + status = TRUE; + *confidence = match_confidence; + lines = g_strsplit(tmpbuf->str, line_termination, 0); + for (line_idx = 0; status && (line = lines[line_idx]); line_idx++) { + rdptr = strstr(line, comment_leader); + if (rdptr) + *rdptr = '\0'; + line = g_strstrip(line); + if (!line || !*line) + continue; + cols = g_strsplit(line, column_separator, 0); + if (!cols) { + status = FALSE; + break; + } + for (col_idx = 0; status && (col = cols[col_idx]); col_idx++) { + if (strspn(col, binary_charset) != strlen(col)) { + status = FALSE; + break; + } + } + g_strfreev(cols); + } + g_strfreev(lines); + g_string_free(tmpbuf, TRUE); + + if (!status) + return SR_ERR; + + return SR_OK; +} + static int init(struct sr_input *in, GHashTable *options) { struct context *inc; @@ -1128,15 +1270,30 @@ static int init(struct sr_input *in, GHashTable *options) * Check the channel list for consistency across file re-import. See * the VCD input module for more details and motivation. */ +static void release_df_channels(struct context *inc, GSList **l) +{ + size_t idx; + + if (!inc->analog_channels || !l) + return; + for (idx = 0; idx < inc->analog_channels; idx++) + g_slist_free(l[idx]); + g_free(l); +} static void keep_header_for_reread(const struct sr_input *in) { struct context *inc; inc = in->priv; + g_slist_free_full(inc->prev_sr_channels, sr_channel_free_cb); inc->prev_sr_channels = in->sdi->channels; in->sdi->channels = NULL; + + release_df_channels(inc, inc->prev_df_channels); + inc->prev_df_channels = inc->analog_datafeed_channels; + inc->analog_datafeed_channels = NULL; } static int check_header_in_reread(const struct sr_input *in) @@ -1155,10 +1312,15 @@ static int check_header_in_reread(const struct sr_input *in) sr_err("Channel list change not supported for file re-read."); return FALSE; } + g_slist_free_full(in->sdi->channels, sr_channel_free_cb); in->sdi->channels = inc->prev_sr_channels; inc->prev_sr_channels = NULL; + release_df_channels(inc, inc->analog_datafeed_channels); + inc->analog_datafeed_channels = inc->prev_df_channels; + inc->prev_df_channels = NULL; + return TRUE; } @@ -1235,7 +1397,7 @@ static int initial_parse(const struct sr_input *in, GString *buf) ret = SR_ERR; goto out; } - sr_dbg("DIAG Got %zu columns in text line: %s.", num_columns, line); + sr_dbg("Got %zu columns in text line: %s.", num_columns, line); /* * Interpret the user provided column format specs. This might @@ -1285,10 +1447,6 @@ static int initial_parse(const struct sr_input *in, GString *buf) if (inc->analog_channels) { size_t sample_size, sample_count; - size_t detail_idx; - struct column_details *detail; - int *digits_item; - void *channel; sample_size = sizeof(inc->analog_datafeed_buffer[0]); inc->analog_datafeed_buf_size = CHUNK_SIZE; inc->analog_datafeed_buf_size /= sample_size; @@ -1301,17 +1459,6 @@ static int initial_parse(const struct sr_input *in, GString *buf) goto out; } inc->analog_datafeed_buf_fill = 0; - inc->analog_datafeed_channels = g_malloc0(inc->analog_channels * sizeof(inc->analog_datafeed_channels[0])); - inc->analog_datafeed_digits = g_malloc0(inc->analog_channels * sizeof(inc->analog_datafeed_digits[0])); - digits_item = inc->analog_datafeed_digits; - for (detail_idx = 0; detail_idx < inc->column_want_count; detail_idx++) { - detail = &inc->column_details[detail_idx]; - if (!format_is_analog(detail->text_format)) - continue; - channel = g_slist_nth_data(in->sdi->channels, detail->channel_index); - inc->analog_datafeed_channels[detail->channel_offset] = g_slist_append(NULL, channel); - *digits_item++ = detail->analog_digits; - } } out: @@ -1388,7 +1535,8 @@ static int process_buffer(struct sr_input *in, gboolean is_eof) const struct column_details *details; col_parse_cb parse_func; int ret; - char *p, **lines, *line, **columns, *column; + char *processed_up_to; + char **lines, *line, **columns, *column; inc = in->priv; if (!inc->started) { @@ -1412,16 +1560,17 @@ static int process_buffer(struct sr_input *in, gboolean is_eof) if (!in->buf->len) return SR_OK; if (is_eof) { - p = in->buf->str + in->buf->len; + processed_up_to = in->buf->str + in->buf->len; } else { - p = g_strrstr_len(in->buf->str, in->buf->len, inc->termination); - if (!p) - return SR_ERR; - *p = '\0'; - p += strlen(inc->termination); + processed_up_to = g_strrstr_len(in->buf->str, in->buf->len, + inc->termination); + if (!processed_up_to) + return SR_OK; + *processed_up_to = '\0'; + processed_up_to += strlen(inc->termination); } - g_strstrip(in->buf->str); + /* Split input text lines and process their columns. */ ret = SR_OK; lines = g_strsplit(in->buf->str, inc->termination, 0); for (line_idx = 0; (line = lines[line_idx]); line_idx++) { @@ -1498,7 +1647,7 @@ static int process_buffer(struct sr_input *in, gboolean is_eof) g_strfreev(columns); } g_strfreev(lines); - g_string_erase(in->buf, 0, p - in->buf->str); + g_string_erase(in->buf, 0, processed_up_to - in->buf->str); return ret; } @@ -1555,10 +1704,12 @@ static int end(struct sr_input *in) static void cleanup(struct sr_input *in) { - struct context *inc; + struct context *inc, save_ctx; + /* Keep channel references between file re-imports. */ keep_header_for_reread(in); + /* Release dynamically allocated resources. */ inc = in->priv; g_free(inc->termination); @@ -1567,12 +1718,31 @@ static void cleanup(struct sr_input *in) inc->datafeed_buffer = NULL; g_free(inc->analog_datafeed_buffer); inc->analog_datafeed_buffer = NULL; + g_free(inc->analog_datafeed_digits); + inc->analog_datafeed_digits = NULL; + /* analog_datafeed_channels was released in keep_header_for_reread() */ + /* TODO Release channel names (before releasing details). */ + g_free(inc->column_details); + inc->column_details = NULL; + + /* Clear internal state, but keep what .init() has provided. */ + save_ctx = *inc; + memset(inc, 0, sizeof(*inc)); + inc->samplerate = save_ctx.samplerate; + inc->delimiter = save_ctx.delimiter; + inc->comment = save_ctx.comment; + inc->column_formats = save_ctx.column_formats; + inc->start_line = save_ctx.start_line; + inc->use_header = save_ctx.use_header; + inc->prev_sr_channels = save_ctx.prev_sr_channels; + inc->prev_df_channels = save_ctx.prev_df_channels; } static int reset(struct sr_input *in) { - struct context *inc = in->priv; + struct context *inc; + inc = in->priv; cleanup(in); inc->started = FALSE; g_string_truncate(in->buf, 0); @@ -1627,7 +1797,7 @@ static struct sr_option options[] = { }, [OPT_HEADER] = { "header", "Get channel names from first line.", - "Use the first processed line's column captions (when available) as channel names. Off by default", + "Use the first processed line's column captions (when available) as channel names. Enabled by default.", NULL, NULL, }, [OPT_SAMPLERATE] = { @@ -1664,7 +1834,7 @@ static const struct sr_option *get_options(void) l = g_slist_append(l, g_variant_ref_sink(g_variant_new_string("oct"))); options[OPT_SINGLE_FMT].values = l; options[OPT_START_LINE].def = g_variant_ref_sink(g_variant_new_uint32(1)); - options[OPT_HEADER].def = g_variant_ref_sink(g_variant_new_boolean(FALSE)); + options[OPT_HEADER].def = g_variant_ref_sink(g_variant_new_boolean(TRUE)); options[OPT_SAMPLERATE].def = g_variant_ref_sink(g_variant_new_uint64(0)); options[OPT_COL_SEP].def = g_variant_ref_sink(g_variant_new_string(",")); options[OPT_COMMENT].def = g_variant_ref_sink(g_variant_new_string(";")); @@ -1678,7 +1848,9 @@ SR_PRIV struct sr_input_module input_csv = { .name = "CSV", .desc = "Comma-separated values", .exts = (const char*[]){"csv", NULL}, + .metadata = { SR_INPUT_META_FILENAME, SR_INPUT_META_HEADER | SR_INPUT_META_REQUIRED }, .options = get_options, + .format_match = format_match, .init = init, .receive = receive, .end = end,