Only retain pending plots method for plot display

Drop the older method of autoshow, and document the loss of the ability
to suppress plots with semicolons.

NEWS.md

@@ -1,21 +1,25 @@
 # reticulate (development version)
 
 - New optional feature: Reticulate now accepts a new option `jupyter_compat`
-  set to `FALSE` by default, that changes the default display behavior of
-  Reticulate chunks, to better match the behavior of Jupyter.  In the
-  Reticulate default, each standalone code expression in the code chunk that
-  does not end in a semi-colon, generates display of the expression output. For
-  Matplotlib plots, an explicit `plt.show()` generates the figure, but
-  otherwise, the chunk (by default) only generates a figure if the last line in
-  the chunk is an expression returning a recognized Matplotlib object, and
-  without a semi-colon.  With `jupyter_compat=TRUE`, a standalone expression
-  returning a Matplotlib object, anywhere in the chunk, will cause the plot to
-  be displayed automatically after the chunk.  No expression in the chunk will
+  set to `FALSE` by default, that changes the default expression output display
+  behavior of Reticulate chunks, to better match the behavior of Jupyter.  In
+  the Reticulate default, each standalone code expression in the code chunk
+  that does not end in a semi-colon, generates display of the expression
+  output. With the `jupyter_compat` option set, no expression in the chunk will
   generate output, except if there is a standalone expression as the last code
-  statement in the chunk, and that expression does not have a semicolon
-  — a semicolon suppresses the output, as it does for the default behavior.
-  See [PR](https://github.com/rstudio/reticulate/pull/1394) and [original
-  issue](https://github.com/rstudio/reticulate/issues/1391).
+  statement in the chunk, and that expression does not have a semicolon.
+  A semicolon always suppresses the expression output, for the default and
+  `jupyter_compat` case. See
+  [PR](https://github.com/rstudio/reticulate/pull/1394) and [original
+  issue](https://github.com/rstudio/reticulate/issues/1391) for discussion for
+  this and the next item.
+
+- Behavior change: Previously, a Matplotlib plot would only be automatically
+  displayed (without `plt.show()`) if there was a final standalone expression
+  returning a Matplotlib object, and that expression did not have a final
+  semicolon.  With this update, any standalone expression returning
+  a Matplotlib object, with or without a semicolon, will cause chunk to display
+  the plot automatically.  See above for discussion.
 
 - Fix: the knitr engine now automatically calls `plt.show()` for matplotlib 
   bar plots, like it does for other matplotlib plot types (#1391). 

R/knitr-engine.R

@@ -198,7 +198,6 @@ eng_python <- function(options) {
 
     # clear the last value object (so we can tell if it was updated)
     py_compile_eval("'__reticulate_placeholder__'")
-    .engine_context$matplotlib_show_was_called <- FALSE
 
     # run code and capture output
     captured <- if (capture_errors)
@@ -209,8 +208,7 @@ eng_python <- function(options) {
     # handle matplotlib and other plot output
     captured <- eng_python_autoprint(
       captured = captured,
-      options  = options,
-      autoshow = (last_range & !jupyter_compat)
+      options  = options
     )
 
     # In all modes, code statements ending in semicolons always suppress repr
@@ -457,7 +455,6 @@ eng_python_initialize_matplotlib <- function(options, envir) {
   # override show implementation
   plt$show <- function(...) {
 
-    .engine_context$matplotlib_show_was_called <- TRUE
     .engine_context$matplotlib_pending_show = FALSE
 
     # get current chunk options
@@ -588,7 +585,7 @@ eng_python_altair_chart_id <- function(options, ids) {
 
 }
 
-eng_python_autoprint <- function(captured, options, autoshow) {
+eng_python_autoprint <- function(captured, options) {
 
   # bail if no new value was produced by interpreter
   value <- py_last_value()
@@ -607,18 +604,9 @@ eng_python_autoprint <- function(captured, options, autoshow) {
 
   if (eng_python_is_matplotlib_output(value)) {
 
-    # by default, we suppress "side-effect" outputs from matplotlib
-    # objects; only when 'autoshow' is set will we try to render the
-    # associated matplotlib plot
-    #
-    # handle matplotlib output. note that the default hook installed by
-    # reticulate will update the 'pending_plots' item
-    if (autoshow && !.engine_context$matplotlib_show_was_called) {
-      plt <- import("matplotlib.pyplot", convert = TRUE)
-      plt$show()
-    } else if (isTRUE(options$jupyter_compat)) {
-      .engine_context$matplotlib_pending_show <- TRUE
-    }
+    # Handle matplotlib output. Note that the default hook for plt.show
+    # installed by reticulate will update the 'pending_plots' item.
+    .engine_context$matplotlib_pending_show <- TRUE
 
     # Always suppress Matplotlib reprs
     return("")