Same colour for adjacent numeric columns in DB view

[trantor]

Hello again @tstack .
I've been working with this view below

;CREATE VIEW IF NOT EXISTS deletions AS
WITH stable AS (
    SELECT
        zimbra_audit_log.thread,
        zimbra_audit_log.log_line,
        zimbra_audit_log.log_time,
        zimbra_audit_log.rest,
        json_group_object( jget(content, '/key'), jget(content, '/value') ) AS context
    FROM zimbra_audit_log, regexp_capture_into_json( threadcontext, '(?<key>\w+)=(?<value>[^;]+)' )
    GROUP BY log_line
)
SELECT
    log_line,
    log_time,
    cast(coalesce(regexp_match("(MailboxPurge)",thread) == "MailboxPurge",0) as text) as MailboxPurge,
    1+MAX(cap_table.match_index) AS line_sum,
    SUM( 1 + MAX(cap_table.match_index) ) OVER w1 AS user_mid_sum,
    SUM( 1 + MAX(cap_table.match_index) ) OVER w2 AS user_sum,
    SUM( 1 + MAX(cap_table.match_index) ) OVER w3 AS mid_sum,
    jget(stable.context, '/name') AS name,
    CAST(jget(stable.context, '/mid') as text) AS mid,
    rest
    -- rank() OVER w1 AS rank,
    -- count(log_line) OVER w1 AS count,
FROM stable, regexp_capture(rest, '\b[0-9]+\b') AS cap_table
GROUP BY log_line
WINDOW
    w1 AS (
        PARTITION BY jget(stable.context, '/name'), jget(stable.context, '/mid')
        ORDER BY log_line
        RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
    ),
    w2 AS (
        PARTITION BY jget(stable.context, '/name')
        ORDER BY log_line
        RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
    ),
    w3 AS (
        PARTITION BY jget(stable.context, '/mid')
        ORDER BY log_line
        RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
    )
ORDER BY user_mid_sum DESC, name||mid,log_line;

and for some reason the adjacent columns user_mid_sum and user_sum are shown with the same colour, which makes the colouring slightly useless.

Is this a bug by any chance?

[trantor]

And another thing: is there a way to exclude fields from a view or a mere SELECT to be included among those available for colouring? I've used CAST(... AS TEXT) on a couple of fields, but it's not really ideal.

[tstack]

I've used CAST(... AS TEXT) on a couple of fields, but it's not really ideal.

Yes, CASTing as TEXT is the only way (I think) to keep a column from being charted.

[trantor]

Ok, I was a bit annoyed by the fact the text is left-justified by default. Is that a choice by lnav or the SQLite output?
Another question: by default the view starts by charting all columns; is there a configuration switch, by any chance, to default to charting no column as the starting view?

[tstack]

Ok, I was a bit annoyed by the fact the text is left-justified by default.

The padding in the cells is done in here:

lnav/src/db_sub_source.cc

Lines 62 to 79 in ca4e61b

	auto actual_col_size
	= std::min(MAX_COLUMN_WIDTH, this->dls_headers[lpc].hm_column_size);
	auto cell_str = scrub_ws(this->dls_rows[row][lpc]);
	truncate_to(cell_str, MAX_COLUMN_WIDTH);
	auto cell_length
	= utf8_string_length(cell_str).unwrapOr(actual_col_size);
	auto padding = actual_col_size - cell_length;
	this->dls_cell_width[lpc] = cell_str.length() + padding;
	if (this->dls_headers[lpc].hm_column_type != SQLITE3_TEXT) {
	label_out.append(padding, ' ');
	}
	label_out.append(cell_str);
	if (this->dls_headers[lpc].hm_column_type == SQLITE3_TEXT) {
	label_out.append(padding, ' ');
	}
	label_out.append(1, ' ');

Another question: by default the view starts by charting all columns; is there a configuration switch, by any chance, to default to charting no column as the starting view?

There's no configuration option. I would prefer to keep the configuration options to a minimum and instead find a better UX.

[trantor]

Mmm. Ok for the justification/padding: it makes sense and I can live with it.
I think that rather than just TAB to cycle through the numeric columns, which may be more than a few, it would be a good thing for an additional keybinding which toggles on and off the whole thing, since it would be quite annoying to cycle back and forth.
I am not sure whether the cycling should be rather (I) no column (II) all columns (III...) cycle one column at a time.
Leaving all columns as the initial option allows to "discover" the feature automatically, but it often charts useless columns as well generating noise... I am wondering which is better myself, really.

[tstack]

Yeah, the coloring is an issue. You can press TAB to cycle through the columns and only highlight one at a time

[trantor]

Yep, I know about TAB for the column selection: I was wondering whether there was a way to exclude them completely from the process. I believe you can do that when using a format.