X-Git-Url: https://sigrok.org/gitweb/?a=blobdiff_plain;f=src%2Finput%2Fcsv.c;h=dabd52c86574a122d62f4159d737bd798614c83e;hb=05719d75aa31627a25375567b3641d8787423567;hp=0fe8826611fde2b1c3d6584c7f7ed19b08fc88fa;hpb=9e7af34eaf466a432d5fb02eacd3bb818cc839fc;p=libsigrok.git diff --git a/src/input/csv.c b/src/input/csv.c index 0fe88266..dabd52c8 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" @@ -41,9 +43,14 @@ * "all remaining columns", only applicable to the last field), a format * specifying character ('x' hexadecimal, 'o' octal, 'b' binary, 'l' * single-bit logic), and an optional bit count (translating to: logic - * channels communicated in that column). This "column_formats" option - * is most versatile, other forms of specifying the column layout only - * exist for backwards compatibility. + * channels communicated in that column). The 'a' format marks analog + * data, an optionally following number is the digits count (resolution). + * The 't' format marks timestamp values, which could help in automatic + * determination of the input stream's samplerate. This "column_formats" + * option is most versatile, other forms of specifying the column layout + * only exist for backwards compatibility, and are rather limited. They + * exclusively support logic input data in strictly adjacent columns, + * with further constraints on column layout for multi-bit data. * * single_column: Specifies the column number which contains the logic data * for single-column mode. All logic data is taken from several bits @@ -109,16 +116,36 @@ * - ... -I csv:start_line=20:header=yes:... * Skip the first 19 text lines. Use line 20 to derive channel names. * Data starts at line 21. + * - ... -I csv:column_formats=*a6 ... + * Each column contains an analog value with six significant digits + * after the decimal period. + * - ... -I csv:column_formats=t,2a ... + * The first column contains timestamps, the next two columns contain + * analog values. The capture's samplerate could get determined from + * the timestamp values if not provided by the user by means of the + * 'samplerate' option. This assumes a mere number in units of seconds, + * 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). */ /* * TODO * - * - Add support for analog input data? (optional) - * - Extend the set of supported column types. Just grab a double - * value from floating point format input text. - * - Optionally get precision ('digits') from the column's format spec? - * From the position which is "bit count" for logic channels? + * - 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? @@ -128,12 +155,16 @@ * samplerates, etc). */ +typedef float csv_analog_t; /* 'double' currently is flawed. */ + /* Single column formats. */ enum single_col_format { FORMAT_NONE, /* Ignore this column. */ FORMAT_BIN, /* Bin digits for a set of bits (or just one bit). */ FORMAT_HEX, /* Hex digits for a set of bits. */ FORMAT_OCT, /* Oct digits for a set of bits. */ + FORMAT_ANALOG, /* Floating point number for an analog channel. */ + FORMAT_TIME, /* Timestamps. */ }; static const char *col_format_text[] = { @@ -141,6 +172,8 @@ static const char *col_format_text[] = { [FORMAT_BIN] = "binary", [FORMAT_HEX] = "hexadecimal", [FORMAT_OCT] = "octal", + [FORMAT_ANALOG] = "analog", + [FORMAT_TIME] = "timestamp", }; static const char col_format_char[] = { @@ -148,24 +181,50 @@ static const char col_format_char[] = { [FORMAT_BIN] = 'b', [FORMAT_HEX] = 'x', [FORMAT_OCT] = 'o', + [FORMAT_ANALOG] = 'a', + [FORMAT_TIME] = 't', }; +static gboolean format_is_ignore(enum single_col_format fmt) +{ + return fmt == FORMAT_NONE; +} + +static gboolean format_is_logic(enum single_col_format fmt) +{ + return fmt >= FORMAT_BIN && fmt <= FORMAT_OCT; +} + +static gboolean format_is_analog(enum single_col_format fmt) +{ + return fmt == FORMAT_ANALOG; +} + +static gboolean format_is_timestamp(enum single_col_format fmt) +{ + return fmt == FORMAT_TIME; +} + struct column_details { size_t col_nr; enum single_col_format text_format; size_t channel_offset; size_t channel_count; + int analog_digits; + GString **channel_names; }; struct context { gboolean started; - /* Current selected samplerate. */ + /* Current samplerate, optionally determined from input data. */ uint64_t samplerate; + double prev_timestamp; gboolean samplerate_sent; - /* Number of logic channels. */ + /* Number of channels. */ size_t logic_channels; + size_t analog_channels; /* Column delimiter (actually separator), comment leader, EOL sequence. */ GString *delimiter; @@ -190,10 +249,17 @@ struct context { size_t sample_unit_size; /**!< Byte count for a single sample. */ uint8_t *sample_buffer; /**!< Buffer for a single sample. */ + csv_analog_t *analog_sample_buffer; /**!< Buffer for one set of analog values. */ uint8_t *datafeed_buffer; /**!< Queue for datafeed submission. */ size_t datafeed_buf_size; size_t datafeed_buf_fill; + /* "Striped" layout, M samples for N channels each. */ + csv_analog_t *analog_datafeed_buffer; /**!< Queue for analog datafeed. */ + size_t analog_datafeed_buf_size; + size_t analog_datafeed_buf_fill; + int *analog_datafeed_digits; + GSList **analog_datafeed_channels; /* Current line number. */ size_t line_number; @@ -215,8 +281,32 @@ struct context { * (when it is full, or upon EOF). */ +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); + inc->samplerate_sent = TRUE; + } + + return SR_OK; +} + static void clear_logic_samples(struct context *inc) { + if (!inc->logic_channels) + return; inc->sample_buffer = &inc->datafeed_buffer[inc->datafeed_buf_fill]; memset(inc->sample_buffer, 0, inc->sample_unit_size); } @@ -241,9 +331,6 @@ static int flush_logic_samples(const struct sr_input *in) { struct context *inc; struct sr_datafeed_packet packet; - struct sr_datafeed_meta meta; - struct sr_config *src; - uint64_t samplerate; struct sr_datafeed_logic logic; int rc; @@ -251,17 +338,9 @@ static int flush_logic_samples(const struct sr_input *in) if (!inc->datafeed_buf_fill) return SR_OK; - if (inc->samplerate && !inc->samplerate_sent) { - packet.type = SR_DF_META; - packet.payload = &meta; - samplerate = inc->samplerate; - src = sr_config_new(SR_CONF_SAMPLERATE, g_variant_new_uint64(samplerate)); - meta.config = g_slist_append(NULL, src); - sr_session_send(in->sdi, &packet); - g_slist_free(meta.config); - sr_config_free(src); - inc->samplerate_sent = TRUE; - } + rc = flush_samplerate(in); + if (rc != SR_OK) + return rc; memset(&packet, 0, sizeof(packet)); memset(&logic, 0, sizeof(logic)); @@ -297,6 +376,99 @@ static int queue_logic_samples(const struct sr_input *in) return SR_OK; } +static void set_analog_value(struct context *inc, size_t ch_idx, csv_analog_t value); + +static void clear_analog_samples(struct context *inc) +{ + size_t idx; + + if (!inc->analog_channels) + return; + inc->analog_sample_buffer = &inc->analog_datafeed_buffer[inc->analog_datafeed_buf_fill]; + for (idx = 0; idx < inc->analog_channels; idx++) + set_analog_value(inc, idx, 0.0); +} + +static void set_analog_value(struct context *inc, size_t ch_idx, csv_analog_t value) +{ + if (ch_idx >= inc->analog_channels) + return; + if (!value) + return; + inc->analog_sample_buffer[ch_idx * inc->analog_datafeed_buf_size] = value; +} + +static int flush_analog_samples(const struct sr_input *in) +{ + struct context *inc; + struct sr_datafeed_packet packet; + struct sr_datafeed_analog analog; + struct sr_analog_encoding encoding; + struct sr_analog_meaning meaning; + struct sr_analog_spec spec; + csv_analog_t *samples; + size_t ch_idx; + int digits; + int rc; + + inc = in->priv; + if (!inc->analog_datafeed_buf_fill) + return SR_OK; + + rc = flush_samplerate(in); + if (rc != SR_OK) + return rc; + + samples = inc->analog_datafeed_buffer; + for (ch_idx = 0; ch_idx < inc->analog_channels; ch_idx++) { + digits = inc->analog_datafeed_digits[ch_idx]; + sr_analog_init(&analog, &encoding, &meaning, &spec, digits); + memset(&packet, 0, sizeof(packet)); + packet.type = SR_DF_ANALOG; + packet.payload = &analog; + analog.num_samples = inc->analog_datafeed_buf_fill; + analog.data = samples; + analog.meaning->channels = inc->analog_datafeed_channels[ch_idx]; + analog.meaning->mq = 0; + analog.meaning->mqflags = 0; + analog.meaning->unit = 0; + analog.encoding->unitsize = sizeof(samples[0]); + analog.encoding->is_signed = TRUE; + analog.encoding->is_float = TRUE; +#ifdef WORDS_BIGENDIAN + analog.encoding->is_bigendian = TRUE; +#else + analog.encoding->is_bigendian = FALSE; +#endif + analog.encoding->digits = spec.spec_digits; + rc = sr_session_send(in->sdi, &packet); + if (rc != SR_OK) + return rc; + samples += inc->analog_datafeed_buf_size; + } + + inc->analog_datafeed_buf_fill = 0; + return SR_OK; +} + +static int queue_analog_samples(const struct sr_input *in) +{ + struct context *inc; + int rc; + + inc = in->priv; + if (!inc->analog_channels) + return SR_OK; + + inc->analog_datafeed_buf_fill++; + if (inc->analog_datafeed_buf_fill == inc->analog_datafeed_buf_size) { + rc = flush_analog_samples(in); + if (rc != SR_OK) + return rc; + } + return SR_OK; +} + /* Helpers for "column processing". */ static int split_column_format(const char *spec, @@ -312,6 +484,7 @@ static int split_column_format(const char *spec, /* Get the (optional, decimal, default 1) column count. Accept '*'. */ endp = NULL; if (*spec == '*') { + /* Workaround, strtoul("*") won't always yield expected endp. */ count = 0; endp = (char *)&spec[1]; } else { @@ -328,7 +501,7 @@ static int split_column_format(const char *spec, /* Get the (mandatory, single letter) type spec (-/xob/l). */ format_char = *spec++; switch (format_char) { - case '-': /* Might conflict with number-parsing. */ + case '-': case '/': format_char = '-'; format_code = FORMAT_NONE; @@ -343,6 +516,12 @@ static int split_column_format(const char *spec, case 'l': format_code = FORMAT_BIN; break; + case 'a': + format_code = FORMAT_ANALOG; + break; + case 't': + format_code = FORMAT_TIME; + break; default: /* includes NUL */ return SR_ERR_ARG; } @@ -355,8 +534,8 @@ static int split_column_format(const char *spec, if (!endp) return SR_ERR_ARG; if (endp == spec) - count = 1; - if (format_char == '-') + count = format_is_analog(format_code) ? 3 : 1; + if (format_is_ignore(format_code)) count = 0; if (format_char == 'l') count = 1; @@ -376,21 +555,23 @@ static int make_column_details_from_format(const struct sr_input *in, { struct context *inc; char **formats, *format; - size_t format_count, column_count, bit_count; + size_t format_count, column_count, logic_count, analog_count; size_t auto_column_count; - size_t format_idx, c, b, column_idx, channel_idx; + size_t format_idx, c, b, column_idx, channel_idx, analog_idx; enum single_col_format f; struct column_details *detail; GString *channel_name; size_t create_idx; 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); @@ -402,7 +583,7 @@ static int make_column_details_from_format(const struct sr_input *in, g_strfreev(formats); return SR_ERR_ARG; } - column_count = bit_count = 0; + column_count = logic_count = analog_count = 0; auto_column_count = 0; for (format_idx = 0; format_idx < format_count; format_idx++) { format = formats[format_idx]; @@ -424,12 +605,15 @@ static int make_column_details_from_format(const struct sr_input *in, c = auto_column_count; } column_count += c; - bit_count += c * b; + if (format_is_analog(f)) + analog_count += c; + else if (format_is_logic(f)) + logic_count += c * b; } - sr_dbg("Column format %s -> %zu columns, %zu logic channels.", - column_format, column_count, bit_count); + 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.", @@ -437,8 +621,12 @@ 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 = 0; + column_idx = channel_idx = analog_idx = 0; channel_name = g_string_sized_new(64); for (format_idx = 0; format_idx < format_count; format_idx++) { /* Process a format field, which can span multiple columns. */ @@ -451,18 +639,27 @@ static int make_column_details_from_format(const struct sr_input *in, detail = &inc->column_details[column_idx++]; detail->col_nr = column_idx; detail->text_format = f; - if (detail->text_format) { + if (format_is_analog(detail->text_format)) { + detail->channel_offset = analog_idx; + detail->channel_count = 1; + detail->analog_digits = b; + analog_idx += detail->channel_count; + } else if (format_is_logic(detail->text_format)) { detail->channel_offset = channel_idx; detail->channel_count = b; - channel_idx += b; - } - sr_dbg("detail -> col %zu, fmt %s, ch off/cnt %zu/%zu", - detail->col_nr, col_format_text[detail->text_format], - detail->channel_offset, detail->channel_count); - if (!detail->text_format) + channel_idx += detail->channel_count; + } else if (format_is_ignore(detail->text_format)) { + /* EMPTY */ continue; + } else { + /* + * Neither logic nor analog data, nor ignore. + * Format was noted. No channel creation involved. + */ + continue; + } /* - * Create channels with appropriate names. Optionally + * Pick most appropriate channel names. Optionally * use text from a header line (when requested by the * user). In the absence of header text, channels are * assigned rather generic names. @@ -478,6 +675,13 @@ static int make_column_details_from_format(const struct sr_input *in, caption = NULL; if (!caption || !*caption) caption = NULL; + /* + * 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); @@ -488,15 +692,38 @@ static int make_column_details_from_format(const struct sr_input *in, g_string_printf(channel_name, "%zu", detail->channel_offset + create_idx); } - sr_channel_new(in->sdi, detail->channel_offset + create_idx, - SR_CHANNEL_LOGIC, TRUE, channel_name->str); + detail->channel_names[create_idx] = g_string_new_len(channel_name->str, channel_name->len); } } } - inc->logic_channels = channel_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; } @@ -636,7 +863,7 @@ static int parse_logic(const char *column, struct context *inc, ch_rem--; set_logic_level(inc, ch_idx + 0, bits & (1 << 0)); break; - case FORMAT_NONE: + default: /* ShouldNotHappen(TM), but silences compiler warning. */ return SR_ERR; } @@ -653,6 +880,144 @@ static int parse_logic(const char *column, struct context *inc, return SR_OK; } +/** + * @brief 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. + * @param[in] details The column processing details. + * + * @retval SR_OK Success. + * @retval SR_ERR Invalid input data (empty, or format error). + * + * This routine modifies the analog values in the current sample set, + * based on the text input and a user provided format spec. + */ +static int parse_analog(const char *column, struct context *inc, + const struct column_details *details) +{ + size_t length; + double dvalue; float fvalue; + csv_analog_t value; + int ret; + + if (!format_is_analog(details->text_format)) + return SR_ERR_BUG; + + length = strlen(column); + if (!length) { + sr_err("Column %zu in line %zu is empty.", details->col_nr, + inc->line_number); + return SR_ERR; + } + if (sizeof(value) == sizeof(double)) { + ret = sr_atod_ascii(column, &dvalue); + value = dvalue; + } else if (sizeof(value) == sizeof(float)) { + ret = sr_atof_ascii(column, &fvalue); + value = fvalue; + } else { + ret = SR_ERR_BUG; + } + if (ret != SR_OK) { + sr_err("Cannot parse analog text %s in column %zu in line %zu.", + column, details->col_nr, inc->line_number); + return SR_ERR_DATA; + } + set_analog_value(inc, details->channel_offset, value); + + return SR_OK; +} + +/** + * @brief 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. + * @param[in] details The column processing details. + * + * @retval SR_OK Success. + * @retval SR_ERR Invalid input data (empty, or format error). + * + * This routine attempts to automatically determine the input data's + * samplerate from text rows' timestamp values. Only simple formats are + * supported, user provided values always take precedence. + */ +static int parse_timestamp(const char *column, struct context *inc, + const struct column_details *details) +{ + double ts, rate; + int ret; + + if (!format_is_timestamp(details->text_format)) + return SR_ERR_BUG; + + /* + * Implementor's notes on timestamp interpretation. Use a simple + * approach for improved maintainability which covers most cases + * of input data. There is not much gain in adding complexity, + * users can easily provide the rate when auto-detection fails. + * - Bail out if samplerate is known already. + * - Try to interpret the timestamp (simple float conversion). + * If conversion fails then clear all previous knowledge and + * bail out (non-fatal, perhaps warn). Silently ignore values + * of zero since those could be silent fails -- assume that + * genuine data contains at least two adjacent rows with useful + * timestamps for the feature to work reliably. Annoying users + * with "failed to detect" messages is acceptable here, since + * users expecting the feature to work should provide useful + * data, and there are easy ways to disable the detection or + * ignore the column. + * - If there is no previous timestamp, keep the current value + * for later reference and bail out. + * - If a previous timestamp was seen, determine the difference + * between them, and derive the samplerate. Update internal + * state (the value automatically gets sent to the datafeed), + * and clear previous knowledge. Subsequent calls will ignore + * following input data (see above, rate is known). + * + * TODO Potential future improvements: + * - Prefer rationals over floats for improved precision and + * reduced rounding errors which result in odd rates. + * - Support other formats ("2 ms" or similar)? + */ + if (inc->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).", + column, inc->line_number); + inc->prev_timestamp = 0.0; + return SR_OK; + } + if (!inc->prev_timestamp) { + sr_dbg("First timestamp value %g in line %zu.", + ts, inc->line_number); + inc->prev_timestamp = ts; + return SR_OK; + } + sr_dbg("Second timestamp value %g in line %zu.", ts, inc->line_number); + ts -= inc->prev_timestamp; + sr_dbg("Timestamp difference %g in line %zu.", + ts, inc->line_number); + if (!ts) { + sr_warn("Zero timestamp difference in line %zu.", + inc->line_number); + inc->prev_timestamp = ts; + return SR_OK; + } + rate = 1.0 / ts; + rate += 0.5; + rate = (uint64_t)rate; + sr_dbg("Rate from timestamp %g in line %zu.", rate, inc->line_number); + inc->samplerate = rate; + inc->prev_timestamp = 0.0; + + return SR_OK; +} + /** * @brief Parse routine which ignores the input text. * @@ -676,8 +1041,129 @@ static const col_parse_cb col_parse_funcs[] = { [FORMAT_BIN] = parse_logic, [FORMAT_OCT] = parse_logic, [FORMAT_HEX] = parse_logic, + [FORMAT_ANALOG] = parse_analog, + [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; @@ -690,16 +1176,13 @@ static int init(struct sr_input *in, GHashTable *options) in->priv = inc = g_malloc0(sizeof(*inc)); single_column = g_variant_get_uint32(g_hash_table_lookup(options, "single_column")); - logic_channels = g_variant_get_uint32(g_hash_table_lookup(options, "logic_channels")); - inc->delimiter = g_string_new(g_variant_get_string( g_hash_table_lookup(options, "column_separator"), NULL)); if (!inc->delimiter->len) { sr_err("Column separator cannot be empty."); return SR_ERR_ARG; } - s = g_variant_get_string(g_hash_table_lookup(options, "single_format"), NULL); if (g_ascii_strncasecmp(s, "bin", 3) == 0) { format = FORMAT_BIN; @@ -711,7 +1194,6 @@ static int init(struct sr_input *in, GHashTable *options) sr_err("Invalid single-column format: '%s'", s); return SR_ERR_ARG; } - inc->comment = g_string_new(g_variant_get_string( g_hash_table_lookup(options, "comment_leader"), NULL)); if (g_string_equal(inc->comment, inc->delimiter)) { @@ -724,13 +1206,9 @@ static int init(struct sr_input *in, GHashTable *options) sr_warn("Comment leader and column separator conflict, disabling comment support."); g_string_truncate(inc->comment, 0); } - inc->samplerate = g_variant_get_uint64(g_hash_table_lookup(options, "samplerate")); - first_column = g_variant_get_uint32(g_hash_table_lookup(options, "first_column")); - inc->use_header = g_variant_get_boolean(g_hash_table_lookup(options, "header")); - inc->start_line = g_variant_get_uint32(g_hash_table_lookup(options, "start_line")); if (inc->start_line < 1) { sr_err("Invalid start line %zu.", inc->start_line); @@ -924,12 +1402,40 @@ static int initial_parse(const struct sr_input *in, GString *buf) * for datafeed submission that is a multiple of the unit size. * Allocate the larger buffer, the "sample buffer" will point * to a location within that large buffer later. + * + * TODO Move channel creation here, and just store required + * parameters in the format parser above? Could simplify the + * arrangement that logic and analog channels get created in + * strict sequence in their respective groups. */ - inc->sample_unit_size = (inc->logic_channels + 7) / 8; - inc->datafeed_buf_size = CHUNK_SIZE; - inc->datafeed_buf_size *= inc->sample_unit_size; - inc->datafeed_buffer = g_malloc(inc->datafeed_buf_size); - inc->datafeed_buf_fill = 0; + if (inc->logic_channels) { + inc->sample_unit_size = (inc->logic_channels + 7) / 8; + inc->datafeed_buf_size = CHUNK_SIZE; + inc->datafeed_buf_size *= inc->sample_unit_size; + inc->datafeed_buffer = g_malloc(inc->datafeed_buf_size); + if (!inc->datafeed_buffer) { + sr_err("Cannot allocate datafeed send buffer (logic)."); + ret = SR_ERR_MALLOC; + goto out; + } + inc->datafeed_buf_fill = 0; + } + + if (inc->analog_channels) { + size_t sample_size, sample_count; + sample_size = sizeof(inc->analog_datafeed_buffer[0]); + inc->analog_datafeed_buf_size = CHUNK_SIZE; + inc->analog_datafeed_buf_size /= sample_size; + inc->analog_datafeed_buf_size /= inc->analog_channels; + sample_count = inc->analog_channels * inc->analog_datafeed_buf_size; + inc->analog_datafeed_buffer = g_malloc0(sample_count * sample_size); + if (!inc->analog_datafeed_buffer) { + sr_err("Cannot allocate datafeed send buffer (analog)."); + ret = SR_ERR_MALLOC; + goto out; + } + inc->analog_datafeed_buf_fill = 0; + } out: if (columns) @@ -1084,6 +1590,7 @@ static int process_buffer(struct sr_input *in, gboolean is_eof) /* Have the columns of the current text line processed. */ clear_logic_samples(inc); + clear_analog_samples(inc); for (col_idx = 0; col_idx < inc->column_want_count; col_idx++) { column = columns[col_idx]; col_nr = col_idx + 1; @@ -1103,6 +1610,7 @@ static int process_buffer(struct sr_input *in, gboolean is_eof) /* Send sample data to the session bus (buffered). */ ret = queue_logic_samples(in); + ret += queue_analog_samples(in); if (ret != SR_OK) { sr_err("Sending samples failed."); g_strfreev(columns); @@ -1157,6 +1665,7 @@ static int end(struct sr_input *in) return ret; ret = flush_logic_samples(in); + ret += flush_analog_samples(in); if (ret != SR_OK) return ret; @@ -1179,6 +1688,8 @@ static void cleanup(struct sr_input *in) inc->termination = NULL; g_free(inc->datafeed_buffer); inc->datafeed_buffer = NULL; + g_free(inc->analog_datafeed_buffer); + inc->analog_datafeed_buffer = NULL; } static int reset(struct sr_input *in) @@ -1197,11 +1708,11 @@ enum option_index { OPT_SINGLE_COL, OPT_FIRST_COL, OPT_NUM_LOGIC, - OPT_FORMAT, - OPT_START, + OPT_SINGLE_FMT, + OPT_START_LINE, OPT_HEADER, - OPT_RATE, - OPT_DELIM, + OPT_SAMPLERATE, + OPT_COL_SEP, OPT_COMMENT, OPT_MAX, }; @@ -1209,52 +1720,52 @@ enum option_index { static struct sr_option options[] = { [OPT_COL_FMTS] = { "column_formats", "Column format specs", - "Specifies text columns data types: comma separated list of [][], with -/x/o/b/l format specifiers.", + "Text columns data types. A comma separated list of [][] items. * for all remaining columns. - ignores columns, x/o/b/l logic data, a (and digits) analog data, t timestamps.", NULL, NULL, }, [OPT_SINGLE_COL] = { "single_column", "Single column", - "Enable single-column mode, exclusively use text from the specified column (number starting at 1).", + "Simple single-column mode, exclusively use text from the specified column (number starting at 1). Obsoleted by 'column_formats=4-,x16'.", NULL, NULL, }, [OPT_FIRST_COL] = { "first_column", "First column", - "Number of the first column with logic data in simple multi-column mode (number starting at 1, default 1).", + "First column with logic data in simple multi-column mode (number starting at 1, default 1). Obsoleted by 'column_formats=4-,*l'.", NULL, NULL, }, [OPT_NUM_LOGIC] = { "logic_channels", "Number of logic channels", - "Logic channel count, required in simple single-column mode, defaults to \"all remaining columns\" in simple multi-column mode. Obsoleted by 'column_formats'.", + "Logic channel count, required in simple single-column mode, defaults to \"all remaining columns\" in simple multi-column mode. Obsoleted by 'column_formats=8l'.", NULL, NULL, }, - [OPT_FORMAT] = { + [OPT_SINGLE_FMT] = { "single_format", "Data format for simple single-column mode.", - "The number format of single-column mode input data: bin, hex, oct.", + "The input text number format of simple single-column mode: bin, hex, oct. Obsoleted by 'column_formats=x8'.", NULL, NULL, }, - [OPT_START] = { + [OPT_START_LINE] = { "start_line", "Start line", "The line number at which to start processing input text (default: 1).", NULL, NULL, }, [OPT_HEADER] = { "header", "Get channel names from first line.", - "Use the first processed line's column captions (when available) as channel names.", + "Use the first processed line's column captions (when available) as channel names. Off by default", NULL, NULL, }, - [OPT_RATE] = { + [OPT_SAMPLERATE] = { "samplerate", "Samplerate (Hz)", - "The input data's sample rate in Hz.", + "The input data's sample rate in Hz. No default value.", NULL, NULL, }, - [OPT_DELIM] = { + [OPT_COL_SEP] = { "column_separator", "Column separator", "The sequence which separates text columns. Non-empty text, comma by default.", NULL, NULL, }, [OPT_COMMENT] = { "comment_leader", "Comment leader character", - "The text which starts comments at the end of text lines.", + "The text which starts comments at the end of text lines, semicolon by default.", NULL, NULL, }, [OPT_MAX] = ALL_ZERO, @@ -1269,16 +1780,16 @@ static const struct sr_option *get_options(void) options[OPT_SINGLE_COL].def = g_variant_ref_sink(g_variant_new_uint32(0)); options[OPT_FIRST_COL].def = g_variant_ref_sink(g_variant_new_uint32(1)); options[OPT_NUM_LOGIC].def = g_variant_ref_sink(g_variant_new_uint32(0)); - options[OPT_FORMAT].def = g_variant_ref_sink(g_variant_new_string("bin")); + options[OPT_SINGLE_FMT].def = g_variant_ref_sink(g_variant_new_string("bin")); l = NULL; l = g_slist_append(l, g_variant_ref_sink(g_variant_new_string("bin"))); l = g_slist_append(l, g_variant_ref_sink(g_variant_new_string("hex"))); l = g_slist_append(l, g_variant_ref_sink(g_variant_new_string("oct"))); - options[OPT_FORMAT].values = l; - options[OPT_START].def = g_variant_ref_sink(g_variant_new_uint32(1)); + 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_RATE].def = g_variant_ref_sink(g_variant_new_uint64(0)); - options[OPT_DELIM].def = g_variant_ref_sink(g_variant_new_string(",")); + 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(";")); } @@ -1290,7 +1801,9 @@ 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,