Go around calling the default process filter (reducing GC churn)

Instead of allocating strings and passing them to the filter, pass
the char buffer to a C function implementing the same logic.

* src/process.c (read_process_output_before_insert)
(read_process_output_after_insert):
New functions, extracted from internal-default-process-filter.
(Finternal_default_process_filter): Use them.
(read_and_insert_process_output): New function.  Use them.
(read_process_output_fast): New variable.
(read_process_output): Use it to choose how to insert (bug#66020).

* etc/NEWS: Mention the change.

1 files changed, 7 insertions, 0 deletions

diff --git a/etc/NEWS b/etc/NEWS
index 2349cc0cacb..efb9b7078b8 100644
--- a/etc/NEWS
+++ b/etc/NEWS
@@ -126,6 +126,13 @@ to your init:
 
     (advice-add 'completion-at-point :after #'minibuffer-hide-completions)
 
+** The defult process filter was rewritten in native code.
+The round-trip through the Lisp function
+'internal-default-process-filter' is skipped when the process filter is
+the default one. It's reimplemented in native code, reducing GC churn.
+
+To undo the change, set 'read-process-output-fast' to nil.
+
 
 * Changes in Emacs 30.1