NEWS.md
Renamed handler_rpushbullet() argument device to devices.
Fixed thinko bug in example("handler_ntfy").
handler_ntfy() for reporting on progress via the Ntfy.sh Messaging Service using the ntfy package.withProgressShiny() could produce an `if (config$max_steps ==
handlers(new_handlers) would return NULL, instead of list(), if there were no prior handlers set.
handler_cli(..., format_done = "...", clear = TRUE) would not render the format_done message, if set.
Now the ‘shiny’ and ‘filesize’ handlers are enabled by default. Previously, they were only enabled in interactive mode, but as these are frequently used also in non-interactive mode, it’s less confusing if they’re always enabled, e.g. Shiny applications are often run via a Shiny servers. These handlers can be disabled by setting R option progressr.enable to FALSE.
Option progressr.intrusiveness.auditory has been renamed to progressr.intrusiveness.audio.
Add handler_rpushbullet() for reporting on progress via the Pushbullet Messaging Service using the RPushbullet package.
Now also ‘beepr’, ‘debug’, ‘filesize’, ‘notifier’, ‘rpushbullet’, ‘shiny’, ‘tkprogressbar’, and ‘winprogressbar’ handlers report on interrupts.
Now progress updates of type “finish” supports also updating the progress state, e.g. you can do p(amount = 1.0, type = "finish") whereas previously you had to do p(amount = 1.0) and then p(type = "finish") resulting in two progress conditions being signaled.
When using multiple progression handlers, it would only be first one that was updated as the progressor completed, whereas any following ones would not receive that last update.
The ‘cli’ handler would output a newline when completed.
The ‘cli’ handler did not handle zero-length progressors resulting in Error in rep(chr_complete, complete_len) : invalid 'times' argument when the progressor completed.
The ‘cli’ handler did not work when the cli package was configured to report on progress via progressr, i.e. when setting options(cli.progress_handlers = "progressr").
with_progress() and without_progress() disables the global progress handler temporarily while running to avoid progress updates being handled twice. Previously, it was, technically, possible to have two different progress handlers intertwined.Add handler_cli() for rendering progress updates via the cli package and its cli::cli_progress_bar().
Now handler_progress() creates a progress progress bar that is always rendered by forcing progress::progress_bar$new(..., force = TRUE).
handler_txtprogressbar() gained support for ANSI-colored char ASCII and Unicode symbols.
Now with_progress() asserts that the number of active “output” sinks is the same on exit as on enter, and that the last one closed is the one that was created. If not, an informative error message is produced.
Now all progress handlers assert that the number of active “output” sinks is the same on exit as on enter.
Code that relied on the superseded crayon package has now been updated to use the cli package.
Using with_progress() while the global progress handler was enabled could result in errors for the cli handler, and possibly for other progression handlers developed in the future. Because of this, with_progress() and without_progress() now disables the global progress handler temporarily while running.
The pbmclapply() handler went from 0 to 100% in one step, because we forgot to set the max:imum value.
When the using a ‘winprogressbar’ or a ‘tkprogressbar’ handler, progression messages updates the label component of the progress panel. Now, it is also possible to update the title component based on progression messages. How the title and label components are updated and from what type of progression message is configured via the new inputs argument. For example, inputs = list(title = "sticky_message", label = "message") causes progression messages to update the label component and sticky ones to update both. For backward compatible reasons, the default is inputs = list(title = NULL, label = "message").
Now the demo function slow_sum() outputs also “sticky” messages.
A progressor that signaled progress beyond 100% prevented any further progressors in the same environment to report on progress.
It was not possible to reuse handlers of type ‘progress’ more than once, because they did not fully reset themselves when finished.
The ‘pbcol’ progression handler did not respect clean = FALSE.
progress() is defunct in order to re-use it for other purpose. It is unlikely that anyone really used this function, but if you did, then use cond <- progression() to create a progression condition and then use withRestart(signalCondition(cond), muffleProgression = function(p) NULL) to signal it.The progressor function created by progressor() no longer “inherit” objects from the calling environment, which would, for instance, result in those objects to be exported to parallel workers together with the progressor function, which in turn would come with large time and memory costs.
progressor() no longer records the call stack for progressions by default, because that significantly increases the size of these condition objects, e.g. instead of being 5 kB it may be 500 kB. If a large number of progress updates are signaled and collected, as done, for instance, by futures, then the memory consumption on the collecting end could become very large. The large sizes would also have a negative impact on the performance in parallelization with futures because of the extra overhead of transferring these extra large conditions from the parallel workers back to the main R session. These issues has been there since progressr 0.7.0 (December 2020). To revert to the previous behavior, use progressor(..., trace = TRUE).
progressor() gained argument trace to control whether or not the call stack should be recorded in each progression condition.
Now print() for progressor functions and progression conditions report also on the size of the object, i.e. the number of bytes it requires when serialized, for instance, to and from a parallel worker.
parallel::mclapply(). This would give a false impression that progressr updates would work when using parallel::mclapply(), which is not true. Note however, that it does indeed work when using the future ‘multicore’ backend, which uses forks.Creating a new progressor() will now automatically finish an existing progressor, if one was previously created in the same environment. The previous behavior was to give an error (see below bug fix).
R_PROGRESSR_* environment variables are now only read when the progressr package is loaded, where they set the corresponding progressr.* option. Previously, some of these environment variables were queried by different functions as a fallback to when an option was not set. By only parsing them when the package is loaded, it decrease the overhead in functions, and it clarifies that options can be changed at runtime whereas environment variables should only be set at startup.
When using withProgressShiny(), progression messages now updates the detail component of the Shiny progress panel. Previously, it updated the message component. This can be configured via new inputs argument.
withProgressShiny() gained argument inputs, which can be used to control whether or not Shiny progress components message and detail should be updated based on the progression message, e.g. inputs = list(message = "sticky_message", detail = "message") will cause progression messages to update the detail component and sticky ones to update both.
Now supporting zero-length progressors, e.g. p <- progressor(along = x) where length(x) == 0.
Add handlers("rstudio") to report on progress in the RStudio Console via the RStudio Job interface.
p(amount = 2), it is now possible to also specify the absolute amount of progress made this far, e.g. p(step = 42). Argument amount has not effect when argument step is specified. WARNING: Argument step should only be used when in full control of the order when this progression condition is signaled. For example, it must not be signaled as one of many parallel progress updates signaled concurrently, because we cannot guarantee the order these progressions arrive.local() call would result in: “Error in assign(”…progressor”, value = fcn, envir = envir) : cannot change value of locked binding for …progressor.”progress() is deprecated in order to re-use it for other purpose. It is unlikely that anyone really used this function, but if you did, then use cond <- progression() to create a progression condition and then use withRestart(signalCondition(cond), muffleProgression = function(p) NULL) to signal it.The user can now use handlers(global = TRUE) to enable progress reports everywhere without having to use with_progress(). This only works in R (>= 4.0.0) because it requires global calling handlers.
with_progress() now reports on progress from multiple consecutive progressors, e.g. with_progress({ a <- slow_sum(1:3); b <- slow_sum(1:3) }).
A progressor must not be created in the global environment unless wrapped in with_progress() or without_progress() call. Ideally, a progressor is created within a function or a local() environment.
Package now requires R (>= 3.5.0) in order to protect against interrupts.
progressor() gained argument enable to control whether or not the progressor signals progression conditions. It defaults to option progressr.enable so that progress updates can be disabled globally. The enable argument makes it easy for package developers who already provide a progress = TRUE/FALSE argument in their functions to migrate to the progressr package without having to change their existing API, e.g. the setup becomes p <- progressor(along = x, enabled = progress). The p() function created by p <- progressor(..., enable = FALSE) is an empty function with near-zero overhead.
Now with_progress() and without_progress() returns the value of the evaluated expression.
The progression message can now be created dynamically based on the information in the progression condition. Specifically, if message is a function, then that function will called with the progression condition as the first argument. This function should return a character string. Importantly, it is only when the progression handler receives the progression update and calls conditionMessage(p) on it that this function is called.
progressor() gained argument message to set the default message of all progression updates, unless otherwise specified.
progressor() gained argument on_exit = TRUE.
Now the progress handler shows also a spinner by default.
Add the ‘pbcol’ handler, which renders the progress as a colored progress bar in the terminal with any messages written in the front.
Progression handlers now return invisibly whether or not they are finished.
Zero-amount progress updates never reached the progress handlers.
Argument enable for with_progress() had no effect.
with_progress() makes sure that any output produced while reporting on progress will not interfere with the progress output and vice versa, which otherwise is a common problem with progress frameworks that output to the terminal, e.g. progress-bar output is interweaved with printed objects. In contrast, when using progressr we can use message() and print() as usual regardless of progress being reported or not.Signaling progress(msg, class = "sticky") will cause the message to be sticky, e.g. for progress bars outputting to the terminal, the message will be “pushed” above the progress bar.
with_progress() gained argument delay_terminal whose default will be automatically inferred from inspecting the currently set handlers and whether they output to the terminal or not.
Arguments delay_stdout and delay_conditions for with_progress() is now agile to the effective value of the delay_terminal argument.
Now handler_nnn() functions pass additional arguments in ... to the underlying progress-handler backend, e.g. handler_progress(width = 40L) will set up progress::progress_bar$new(width = 40L).
Add environment variables R_PROGRESSR_CLEAR, R_PROGRESSR_ENABLE, R_PROGRESSR_ENABLE_AFTER, R_PROGRESSR_TIMES, and R_PROGRESSR_INTERVAL for controlling the default value of the corresponding progressr.* options.
Limiting the frequency of progress reporting via handler arguments times, interval or intrusiveness did not work and was effectively ignored.
The progress handler, which uses progress::progress_bar(), did not support colorization of the format string when done by the crayon package.
handlers() did not return invisible (as documented).
Argument target was ignored for all handler functions.
Argument interval was ignored for handler_debug().
The class of handler_<nnn>() functions where all reset_progression_handler rather than <nnn>_progression_handler. The same bug caused the reported name field to be "reset" rather than "<nnn>".
progressor() gained arguments offset and scale, and transform.
handlers() gained argument append to make it easier to append handlers.
winprogressbar_handler() would produce error “invalid ‘Label’ argument”.
handlers() did not return a list if the ‘default’ handler was returned.
withProgress2() to withProgressShiny().handlers() gained argument default specifying a progression handler to be returned if none is set.Add withProgress2(), which is a plug-in backward compatibility replacement for shiny::withProgress() wrapped in progressr::with_progress() where the the “shiny” progression handler is by default added to the list of progression handlers used.
Add demo("mandelbrot", package = "progressr").
Now it is possible to send “I’m still here” progression updates by setting the progress step to zero, e.g. progress(amount = 0). This type of information can for instance be used to updated a progress bar spinner.
Add utility function handlers() for controlling option progressr.handlers.
Progression handlers’ internal state now has a sticky message field, which hold the most recent, non-empty progression message received.
with_progress() gained arguments enable and interval as an alternative to setting corresponding options progressr.*.
Now option progressr.interval defaults to 0.0 (was 0.5 seconds).
Added print() for progression_handler objects.
with_progress(..., delay_conditions = "condition"), introduced in progressr 0.1.0, would also capture conditions produced by progression handlers, e.g. progress::progress_bar() output would not be displayed until the very end.with_progress() now captures standard output and conditions and relay them at then end. This is done in order to avoid interweaving such output with the output produced by the progression handlers. This behavior can be controlled by arguments delay_stdout and delay_condition.Add print() for progression conditions and progressor functions.
Now the progressors record more details on the session information. This information is passed along with all progression conditions as part of the internal owner information.
Add filesize_handler progression handler.
Add support for times = 1L for progression handlers which when used will cause the progression to only be presented upon completion (= last step).
The shutdown control_progression signaled by with_progress() on exit now contains the status of the evaluation. If the evaluation was successful, then status = "ok", otherwise "incomplete". Examples of incomplete evaluations are errors and interrupts.
Add utils::winProgressBar() progression handler for MS Windows.
Add support for silent sounds for beepr::beep().
Add option progressr.enable, which defaults to interactive().
Add argument enable_after for progression handlers.
Now with_progress(..., cleanup = TRUE) will signal a generic “shutdown” progression at the end that will trigger all progression handlers to finish up regardless of all steps have been take or not.
Now progressions originating from an unknown source are ignored.
The default output format of the progress::progress_bar() progression handler is now ":percent :bar :message".
The tcltk::tkProgressBar() progression handler now displays the progression message.
Now the progression condition itself is passed to the progression reporter functions.
Add ‘debug_handler’ for prototyping and debugging purposes.
Add ‘newline_handler’ to add newlines between output of multiple handlers.
Argument intrusiveness may now be zero. Previously it had to be a strictly positive value.
Add without_progress() - which causes all progression conditions to be muffled and ignored.
Progressor functions could produce progression conditions that had the same identifiers and therefore would be considered duplicates such that progression handlers would ignore them.
It was an error if a progression took a step big enough to skip more than the next milestone.
Progression handlers now keep the internal step field within [0, max_steps] in case of a too big progression step is taken.
Progression updates received after progression handler is finished would keep increasing the internal step field.
consume_progression to muffleProgression to align with restarts muffleMessage and muffleWarning in base R.Add a plyr-compatible “progress bar” named progress_progressr().
Add option progressr.clear.
Visual progression handler will now always render the complete update state when clear is FALSE.
Now progression handlers ignore a re-signaled progression condition if it has already been processed previously.
Now each progression condition holds unique identifiers for the R session and for the progressor that produced the condition. It also contains an unique index per progressor that is incremented whenever a new progression condition is created.
Add argument auto_done to automatically have progress updates also signal “done” as soon as the last step has been reached.
Made amount the first argument of progressors to avoid having to specify it by name if progressing with an amount than the default amount = 1.0.
Add argument clear to control whether progress reporter should clear its output upon completion. The default is to do this, where supported.
Add progress update handler based on pbmcapply::progressBar().
Each achieved step is now timestamped.
Add option progressr.debug.
Add progressor().
Add progress update handlers based on utils::txtProgressBar(), tcltk::tkProgressBar(), cat("\a"), progress::progress_bar(), beepr::beep(), and notifier::notify().
Add with_progress().
Add options progressr.handlers for settings default progress handlers.
Add progressr.times for controlling the number of times progress updates are rendered.
Add progressr.interval for controlling the minimum number of seconds that needs to elapse before reporting on the next update.
Add progress() to create and signal progression condition.
Add progression() to create progression condition.