Warning and message classes

Every warning tidycreel.connect raises has a class, so one can be handled in one call without silencing the others. The classes follow tidycreel's (see tidycreel::creel_warnings): each warning carries its own class, creel_warning_<fact>, and the parent creel_warning; information that is not a warning is raised as a message with a class creel_message_<fact> and the parent creel_message. A fact tidycreel also raises has tidycreel's class, so one handler covers both packages.

Handle one for one call with withCallingHandlers():

withCallingHandlers(
  fetch_interviews(conn),
  creel_message_columns_dropped = function(m) invokeRestart("muffleMessage")
)

Prefer this to suppressWarnings() or suppressMessages(), which also hide the warnings that mean the fetched data are not what the source holds.

Tiers

Tier 1

The fetched data may be wrong for estimation.

Tier 2

Worth a look; no estimate is affected.

Tier 3

Information, raised as a message rather than a warning.

Warnings (tiers 1 and 2) print on every call. The tier-3 notes about the source, creel_message_columns_dropped, creel_message_columns_by_canonical_name and creel_message_corrections_applied, describe the source rather than one fetch, so they print once per connection: the first fetch of a table prints them, later fetches of that table on the same connection do not, unless the columns differ. A new connection prints them again. Other tier-3 messages print on every call. Unlike tidycreel, connect returns data frames rather than designs, so nothing is recorded for tidycreel::get_warnings().

Warnings

creel_warning_dates_coerced_na

Tier 1. Some dates could not be parsed and became NA.

creel_warning_design_length_disagrees

Tier 1. A design table's stored shift length disagrees with its clock window. See fetch_design().

creel_warning_numeric_coerced_na

Tier 1. Some values of a numeric column could not be read as numbers and became NA.

creel_warning_fetch_created_duplicates

Tier 1. The fetch made interview rows exact duplicates, so each copy counts that interview again: usually by dropping unmapped columns of a long-format source, sometimes by coercing or mapping values to the same value.

creel_warning_repeated_sampling_units

Tier 1. Counts have repeated dates with no count time to tell them apart.

creel_warning_tz_option_differs

Tier 1. A profile's tz: differs from the tidycreel.tz option already set, which is kept, so a night design given no tz is read in the option's zone, not the profile's. See creel_connect_from_yaml().

creel_warning_bins_without_count

Tier 1. Length bins were fetched without a fish count, so a distribution built from them is weighted by rows, not fish.

creel_warning_discovery_field_missing

Tier 2. The discovery response lacks an expected field, which is filled with NA.

creel_warning_odbc_drivers_unlisted

Tier 2. creel_check_driver() could not list the ODBC drivers.

Messages (tier 3)

creel_message_argument_ignored

Tier 3. An argument has no effect in this call: schema column mappings on the API backend.

creel_message_columns_by_canonical_name

Tier 3. Unmapped source columns were taken because they already carry a canonical name. Once per connection.

creel_message_columns_dropped

Tier 3. Source columns not mapped are not carried through. Once per connection.

creel_message_corrections_applied

Tier 3. A profile's corrections file changed fetched values; each change is listed. Once per connection. The result also carries them as attr(x, "corrections"). See creel_connect_from_yaml().

creel_message_tz_option_set

Tier 3. A profile's tz: set the tidycreel.tz option, which was unset, for the session.