rlang::abort() inside
of evaluate().Setting ACTIONS_STEP_DEBUG=1 (as in a failing GHA
workflow) will automatically set log_echo and
log_warning to TRUE (#175).
evaluate works on R 3.6.0 once again.
evaluate() improvements:
Now terminates on the first error in a multi-expression input,
i.e. 1;stop('2');3 will no longer evaluate the third
component. This matches console behaviour more closely.
Calls from conditions emitted by top-level code are automatically stripped (#150).
Result has a class
(evaluate_evaluation/list) with a basic print
method.
Plots created before messages/warnings/errors are correctly captured (#28).
Handler improvements:
The default value handler now evaluates
print() in a child of the evaluation environment. This
largely makes evaluate easier to test, but should make defining S3
methods for print a little easier (#192).
The source output handler is now passed the entire
complete input expression, not just the first component.
evalute(include_timing) has been deprecated. I can’t
find any use of it on GitHub, and it adds substantial code complexity
for little gain.
is.value() has been removed since it tests for an
object that evaluate never creates.
New local_reproducible_output() helper that sets
various options and env vars to help ensure consistency of output across
environments.
parse_all() adds a \n to the end of
every line, even the last one if it didn’t have one in the input.
Additionally, it no longer has a default method, which will generate
better errors if you pass in something unexpected.
New trim_intermediate_plots() drops intermediate
plots to reveal the complete/final plot (#206).
watchout() is no longer exported; it’s really an
implementation detail that should never have been leaked to the public
interface.
source output handler can now take two arguments
(the unparsed src and the parsed call) and
choose to affect the displayed source.Prevent existing plots from leaking into evaluate()
results (thanks, @dmurdoch, yihui/knitr#2297).
If the environment variable
R_EVALUATE_BYPASS_MESSAGES is set to true, the arguments
keep_message and keep_warning of
evaluate() will be set to NA, regardless of
user input, which means messages and warnings will not be captured by
evaluate(). This provides a possibility to force logging
messages and warnings (thanks, @slodge, yihui/yihui.org#1458).
plot_calls()
that made the examples of recordGraphics fail to run on its
help page (thanks, Kurt Hornik).evaluate() gains log_echo and
log_warning arguments. When set to TRUE these
cause code and warnings (respectively) to be immediately emitted to
stderr(). This is useful for logging in unattended
environments (#118).
Improved the error message when users accidentally called
closeAllConnections() (thanks, @guslipkin,
quarto-dev/quarto-cli#5214).
keep_message and
keep_warning of evaluate() can take the value
NA now, which means evaluate() will not
capture the messages and they will be sent to the console. This is
equivalent to the FALSE value before v0.19 (thanks, @gadenbuie,
https://github.com/yihui/yihui.org/discussions/1458).In evaluate(), keep_message and
keep_warning will completely drop messages and warnings,
respectively, when their values are FALSE. Previously
messages would still be emitted (to the console) even if they take
FALSE values.
Fixed the bug that parse_all() fails with line
directives (thanks, @ArcadeAntics, #114).
ggplot2::qplot().new_output_handler() gains a
calling_handlers argument. These are passed to
withCallingHandlers() before evaluate()
captures any conditions.
Fixed #106: do not assume that is.atomic(NULL)
returns TRUE (thanks, @mmaechler).
persp, before.plot.new, and
before.grid.newpage set by users will be respected
throughout the R session (thanks, @KKPMW, #96).Fix for regression introduced in 0.10.1 in parse_all.call() (fixes #77)
evaluate() now respects options(warn >= 2); all warnings are turned into errors (#81)
Added option for the evaluate function to include timing information of ran commands. This information will be subsequently rendered by the replay. Example usage: evaluate::replay(evaluate::evaluate(‘Sys.sleep(1)’, include_timing = TRUE))
Added a new function flush_console() to emulate
flush.console() in evaluate() (#61).
Added a inject_funs() function to create functions
in the environment passed to the envir argument of
evaluate().
allow_error to
parse_all() to allow syntactical errors in R source code
when allow_error = TRUE; this means
evaluate(stop_on_error = 0 or 1) will no longer stop on
syntactical errors but returns a list of source code and the error
object instead. This can be useful to show syntactical errors for
pedagogical purposes.filename to evaluate() and
parse_all() (thanks, @flying-sheep, #58).Plots are no longer recorded when the current graphical device has been changed, which may introduce issues like yihui/knitr#824.
parse_all() can parse R code that contains multibyte
characters correctly now (#49, yihui/knitr#988)
Actually use the text and graphics in
new_output_handler
Multiple expressions separated by ; on the same line
can be printed as expected when the result returned is visible,
e.g. both x and y will be printed when the
source code is x; y. In previous versions, only
y is printed. (thanks, Bill Venables)
fixed yihui/knitr#600: when the last expression in the code is a comment, the previous incomplete plot was not captured
the empty plots produced by strwidth(), strheight(), and clip() are no longer recorded
evaluate() no longer records warnings in case of options(warn = -1); see yihui/knitr#610
for ‘output_handler’ in evaluate(), visible values from the ‘value’ handler will be saved to the output list; this makes it possible for users to save the original values instead of their printed side effects; this change will not affect those who use the default output handlers (#40, thanks, Gabriel Becker)
the ‘value’ handler in new_output_handler() may take an additional argument that means if the value is visible or not; this makes it possible to save the invisible values as well (#41, thanks, Joroen Ooms)
fixed #25: plots can be correctly recorded under a complex layout now (#25, thanks, Jack Tanner and Andy Barbour)
fixed yihui/knitr#582: evaluate() misclassified some plot changes as “par changes” and removed some plots when it should not; now it is better at identifying plot changes dur to par() (thanks, Keith Twombley)
Perspective plots from persp() are captured now
(thanks to Harvey Lime and Yihui Xie)
If an error occurs during printing a visible value, evaluate will halt on a cryptic error “operator is invalid for atomic vectors” (#26, fixed by Yihui Xie)
If the internal connection was accidentally closed by the user, a more informative message will show up (#23)
Now the graphical device will always try to record graphics by default (when new_device = TRUE) (#34)
Some empty and incomplete plots caused by par() or layout() will be filtered out correctly for R 3.0 (#35)
Added output_handler argument to
evaluate. Should be a output_handler object,
which is a list of functions for handling each type of result, prior to
printing of visible return values. This allows clients to override the
console-like printing of values, while still processing them in the
correct temporal context. The other handlers are necessary to convey the
correct ordering of the output. This essentially provides stream-based
processing, as an alternative to the existing deferred
processing.
New option, stop_on_error which controls behaviour
when errors occur. The default value, 0, acts like you’ve
copied and pasted the code into the console, and continues to execute
all code. 1 will stop the code execution and return the
results of evaluation up to that point, and 2 will raise an
error.
Compound expressions like x <- 10; x are now
evaluated completely.
Chinese characters on windows now work correctly (thanks to Yihui Xie)
Graphics and output interleaved correctly when generated from a loop or other compound statements
By default, evaluate will now open a new graphics
device and clean it up afterwards. To suppress that behaviour use
new_device = FALSE
use show to display S4 objects.
replace deprecated .Internal(eval.with.vis) with
correct withVisible
evaluate gains debug argument
test_package to avoid problems with latest version
of testthatImport stringr instead of depending on it.
Test plot recording only in the presence of interactive devices.
try_capture_stack and create_traceback do a much better job of removing infrastructure calls from the captured traceback
visible results are automatically evaluated and their outputs are captured. This is particularly important for lattice and ggplot graphics, which otherwise require special handling. It also correctly captures warnings, errors and messages raised by the print method.