with_timeout: Set a time limit for one call
In tidymedia: Media File Preprocessing and Metadata for the 'tidyverse'

with_timeout

R Documentation

Set a time limit for one call

Description

with_timeout() runs expr with a time limit of its own. The limit applies to each FFmpeg, FFprobe or MediaInfo program that expr starts. When with_timeout() returns, or stops with an error, the limit that was in force before the call is back.

The session limit, options(tidymedia.timeout = ), applies to every call in the session. with_timeout() applies to one call. For example, you can give one test conversion five minutes in a session with a one-hour limit.

Usage

with_timeout(expr, seconds)

Arguments

`expr`	An expression. It is run once, where you wrote it, and its value is returned.
`seconds`	A whole number of seconds. `0` means no limit, so `with_timeout(expr, 0)` removes a session limit for one call. A fraction, a negative number, a string or `NULL` gives an error before `expr` runs.

Details

The limit applies to each program, not to the whole call. In a 100-row batch inside with_timeout(expr, 600), each program that a row starts gets 600 seconds, plus the delay in "How long the wait can be". The workers of a parallel = TRUE run use the same limit.

The limit is a whole number of seconds. The package does not round a fraction, because R would read a limit below one second as no limit.

A limit set with options(tidymedia.timeout = ) follows the same rule, with one difference. options(tidymedia.timeout = NULL) removes the option, so it means no limit. A function that can start a program gives an error for a wrong value, even when run = FALSE. ffm_batch() gives that error before it starts any job. A function that starts no program gives no such error. For example, has_hardware_encoder() starts none when you set tidymedia.hardware_encoders. A ⁠probe_*()⁠ function that you give a probe object also starts none. has_encoder() is the exception: it checks the limit first, even on a machine where it finds no FFmpeg to start.

Most functions check their own arguments before the limit. So a wrong argument gives its own error, even when the limit is also wrong. A few arguments of the ⁠_batch⁠ functions are checked inside each job, after the limit. An example is the pixel_format of anonymize_video_batch(). When the limit is also wrong, the error is about the limit.

Value

The value of expr.

How long the wait can be

The limit sets how long R waits for a program, and the wait can be longer. When the limit is reached, R asks the program to stop. R asks again 20 seconds later, and kills the program 20 seconds after that. So R can wait up to 40 seconds past the limit. For example, five hung files under a 1-second limit can take about three and a half minutes.

R does not guarantee that the program stops. A program can survive the attempts to stop it. How fast a program stops also depends on its version.

What happens when the limit is reached

A reached limit is never silent. The call gives an error or a warning.

These functions give an error with the class tidymedia_timeout, which names the program and the limit:

the task functions whose names do not end in ⁠_batch⁠, except segment_video()
ffm_run(), ffmpeg(), ffprobe() and mediainfo()
ffmpeg_codecs() and ffmpeg_encoders()
has_encoder() and has_hardware_encoder() when they ask FFmpeg
verify_media(), because a check with no answer is not a "no"

These functions give a warning instead, so that one hung file does not lose the rest of the work:

probe_all(), the other ⁠probe_*()⁠ functions, mediainfo_parameter(), mediainfo_query(), mediainfo_template() and the ⁠get_*()⁠ functions give NA for that file. One warning at the end says how many files timed out.
ffm_batch(), segment_video() and the ⁠_batch⁠ task functions set success = FALSE for that job. One warning at the end says how many jobs timed out. It has the class tidymedia_batch_timeout. Two steps of these calls give an error instead. One is the analysis pass of normalize_audio_batch(two_pass = TRUE). The other is the check that FFmpeg has the hardware encoder that hardware names, such as "nvenc". That check asks FFmpeg only when tidymedia.hardware_encoders is not set and the session has no stored answer. The glossary in vignette("tidymedia") explains hardware encoders.
The dropped-track check of extract_audio(), convert_audio(), normalize_audio() and their ⁠_batch⁠ forms warns that it could not check. The track count that separate_audio_video() reads after a failed run warns the same way. So does the check of FFmpeg's encoder lists that ffm_run(), the task functions and ffm_batch() make after a failed run, once per call (see "When the build lacks an encoder" on ffm_run()). A batch manifest, see ffm_manifest(), and program_status() warn when they cannot read a program version. These warnings have the class tidymedia_probe_timeout, and the call goes on as it would for an unreadable input.

suppressWarnings(classes = "tidymedia_dropped_audio") hides the dropped-track warning, but not the warning that the check timed out. To hide both, add "tidymedia_probe_timeout" to classes.

The task functions and ffm_run() delete a part-written output file after a timeout, as they do after any failed run. ffmpeg() cannot tell which of your arguments is the output, so it leaves that file. Check the output of a timed-out ffmpeg() call yourself.

Examples

# Inside the call, the limit is the one you gave.
with_timeout(getOption("tidymedia.timeout"), 30)

# Outside it, the session's own setting is untouched.
getOption("tidymedia.timeout", default = "unset")


# Bound one conversion at five minutes, whatever the session is set to.
# Needs FFmpeg, and writes to a temporary file.
if (nzchar(Sys.which("ffmpeg"))) {
  video <- system.file("extdata", "sample.mp4", package = "tidymedia")
  with_timeout(extract_audio(video, tempfile(fileext = ".wav")), 300)
}

tidymedia documentation built on Oct. 11, 2026, 5:08 p.m.