loggit2 2.4.0

New Features

set_call_options() offers new option to log condition calls or even the full call stack. get_call_options() returns the current settings.
All condition log handlers e.g. warning() and with_loggit() now support call logging.
Added debuginfo() as an additional log handler for debugging purposes.

loggit() now checks the ... arguments for use of reserved names, currently log_call and timestamp.

Some small changes to increase logging speed
wìth_loggit() now opens a connection to the log file at the beginning of the block and closes it at the end. This increases the speed of logging noticeably, especially for many log entries.
The log file will be created during logging, but will automatically be removed if no log entries were made.
conditionMessage() is now used to extract the message from conditions. This is more robust than con[["message"]] and should work for all conditions, especially also for rlang conditions.

Custom sanitizers and unsanitizers are no longer supported. This decision was made because no active user is known, and this functionality severely limits further development. If custom sanitizers were previously used, they can simply be executed before or after loggit() or read_logs().
If custom sanitizers were used to circumvent bugs, please report them, so they can be fixed.
Special characters are no longer escaped by replacement, but rather by "\".

Added convert_to_csv() to convert log files to CSV format.
Added with_loggit() to log third-party code or to easily use different loggit() parameters for a chunk of code.
NAs are now stored as null in the JSON log, and read_logs() also restores these as NA. This was previously (unintentionally) guaranteed by replacing the NA with "__NA__".
read_logs() now allows reading the logs in reverse order using the last_first argument.
A global log level can now be set using set_log_level(), which is used by all functions unless otherwise stated. The log levels are: "DEBUG", "INFO", "WARN", "ERROR", and "NONE".
Added set_echo() to control globally whether log messages are echoed to the console.
All condition log handlers (e.g., warning()) allow NA for the parameter .loggit. If NA, the log level set by set_log_level() is used to determine if the condition should be logged. This is the new default behavior, but since the default log level is "DEBUG", this should not change the behavior of existing code.
All set_* functions now return the previous value of the setting.
The default settings can now be controlled by the system environment variables FILE_LOGGIT2, TIMESTAMP_LOGGIT2, ECHO_LOGGIT2 and LEVEL_LOGGIT2.

read_logs() now correctly reads empty character values "", as in {"key": ""}. Previously, empty fields were read as NA. This meant that when rotate_logs() was used, these entries could completely disappear from the respective JSON object.
loggit() now does not unintentionally “repair” argument names of log entries. Previously, the names were replaced by check.names of data.frame(), which could lead to unexpected behavior. Names that are not valid JSON keys are now escaped according to the JSON standard.

read_logs() now returns a data.frame with the empty character columns “timestamp”, “log_lvl”, and “log_msg” instead of an empty (0x0) data.frame if the log file has no entries.
The JSON reading functions are more tolerant of manual changes to the log.
The parameter exprs was added to stopifnot() and included in the documentation. This has no impact on functionality due to the specific way base::stopifnot() is called internally.
loggit() now throws an error if there are unnamed ... arguments. Previously, these were silently named by fix.empty.names of data.frame, which could lead to unexpected behavior.
loggit() now also checks the length of the log_lvl and log_msg arguments and only uses the first element. Previously, the log entry had been multiplied, leading to unintended consequences regarding custom_log_lvl.

write_ndjson no longer warns if the log contains unsanitized line breaks. This warning could only be generated by package-internal errors (therefore nonsensical in the CRAN package) or by a custom sanitizer, but in this case only this one character was specifically tested and thus provides a false sense of security.
The package now requires compilation. This is necessary because the JSON parser was written in C++ for faster reading.
loggit2 now requires at least R 4.0.0.

Custom sanitizers and unsanitizers must be able to process character vectors. Previously, only the processing of vectors of length one was explicitly required.

message(), warning(), and stop() now accept conditions as input like their base R equivalents.
New stopifnot() handler.
set_logfile() has a new argument create that allows the user to create the file if it does not exist.

read_logs() processes entries with ": " correctly. Previously, entries were truncated and/or assigned to incorrect columns.
message(), warning(), and stop() now use the same call in their messages and their condition objects as their base R equivalents and no longer give themselves as the call. For warning() and stop(), there can be deviations in very rare cases, as the function that determines the call for these in base R is not provided at the R or C level, nor is the necessary C header.
set_logfile() now attempts to convert relative paths to absolute paths. This prevents the logfile from being inadvertently changed when switching (even temporarily) the working directory. If this unintended side effect was used intentionally, the same effect can be achieved by explicitly resetting the path.
set_logfile() now correctly outputs the randomly generated temporary file as the new path in its confirmation message when NULL is given as an argument.
rotate_logs(rotate_lines = 0L) now empties the log as expected. Additionally, an error is thrown for negative values.
rotate_logs() preserves the original sanitization of the log entries. Previously, the sanitization was lost when the log was rotated for values between 0 and the number of log entries.
read_logs() now correctly reads empty log fields as NA_character_. Previously, empty fields were read as "" when the first entry of the field was empty.

All set_* functions use message() instead of print() for confirmation. This ensures that the confirmations no longer interfere with the log via echo.
default_ndjson_sanitizer(), which was not exported but is visible in the documentation, now follows the rules for sanitizers.

Functions that were only for internal use (and were not exported) were marked as such and are no longer visible in the index.