bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Jim Porter]

[-- Attachment #1: Type: text/plain, Size: 442 bytes --]

Steps to reproduce:

   emacs -Q -f eshell
   $ sleep 100 &
   $ kill -9 <sleep's process id>

After this, you'll see two Eshell prompts get emitted (one with a "[9]" 
for the exit status of "sleep"). We don't want to emit prompts when 
background processes die though.

This is a regression from Emacs 29, likely due to some changes I made to 
improve support for complex background commands. Eli, is this ok to 
merge to the release branch?

[-- Attachment #2: 0001-Don-t-emit-a-prompt-in-Eshell-when-a-background-comm.patch --]
[-- Type: text/plain, Size: 2785 bytes --]

From 3e5a702f0136cc6a071546a854aea54536c87783 Mon Sep 17 00:00:00 2001
From: Jim Porter <jporterbugs@gmail.com>
Date: Tue, 9 Jul 2024 10:45:35 -0700
Subject: [PATCH] Don't emit a prompt in Eshell when a background command is
 killed

* lisp/eshell/esh-cmd.el (eshell-resume-command): Check for
background-ness before resetting the prompt.

* test/lisp/eshell/esh-cmd-tests.el
(esh-cmd-test/background/simple-command): Make the regexp a bit
stricter.
(esh-cmd-test/background/kill): New test.
---
 lisp/eshell/esh-cmd.el            |  7 +++++--
 test/lisp/eshell/esh-cmd-tests.el | 14 +++++++++++++-
 2 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/lisp/eshell/esh-cmd.el b/lisp/eshell/esh-cmd.el
index 0b3137127d2..d0de6e55ce6 100644
--- a/lisp/eshell/esh-cmd.el
+++ b/lisp/eshell/esh-cmd.el
@@ -1040,8 +1040,11 @@ eshell-resume-command
                  (not (string-match eshell-reset-signals status)))
             (eshell-resume-eval command)
           (eshell-remove-command command)
-          (declare-function eshell-reset "esh-mode" (&optional no-hooks))
-          (eshell-reset))))))
+          ;; Reset the prompt if the command we just aborted was in the
+          ;; foreground.
+          (unless (car command)
+            (declare-function eshell-reset "esh-mode" (&optional no-hooks))
+            (eshell-reset)))))))
 
 (defun eshell-resume-eval (command)
   "Destructively evaluate a COMMAND which may need to be deferred.
diff --git a/test/lisp/eshell/esh-cmd-tests.el b/test/lisp/eshell/esh-cmd-tests.el
index 70e1901c169..d8124a19af6 100644
--- a/test/lisp/eshell/esh-cmd-tests.el
+++ b/test/lisp/eshell/esh-cmd-tests.el
@@ -113,7 +113,7 @@ esh-cmd-test/background/simple-command
     (with-temp-eshell
      (eshell-match-command-output
       (format "*echo hi > #<%s> &" bufname)
-      (rx "[echo" (? ".exe") "] " (+ digit) "\n"))
+      (rx bos "[echo" (? ".exe") "] " (+ digit) "\n"))
      (eshell-wait-for-subprocess t))
     (should (equal (buffer-string) "hi\n"))))
 
@@ -129,6 +129,18 @@ esh-cmd-test/background/subcommand
      (eshell-wait-for-subprocess t))
     (should (equal (buffer-string) "olleh\n"))))
 
+(ert-deftest esh-cmd-test/background/kill ()
+  "Make sure that a background command that gets killed doesn't emit a prompt."
+  (skip-unless (executable-find "sleep"))
+  (let ((background-message (rx bos "[sleep" (? ".exe") "] " (+ digit) "\n")))
+    (with-temp-eshell
+      (eshell-match-command-output "*sleep 10 &" background-message)
+      (kill-process (caar eshell-process-list))
+      (eshell-wait-for-subprocess t)
+      ;; Ensure we didn't emit another prompt after killing the
+      ;; background process.
+      (should (eshell-match-output background-message)))))
+
 \f
 ;; Lisp forms
 
-- 
2.25.1

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Eli Zaretskii]

> Date: Tue, 9 Jul 2024 11:04:05 -0700
> From: Jim Porter <jporterbugs@gmail.com>
> 
> This is a regression from Emacs 29, likely due to some changes I made to 
> improve support for complex background commands. Eli, is this ok to 
> merge to the release branch?

I don't think I understand the essence of the change, and thus cannot
appreciate its effects enough to be able to answer this.  What is the
significance of '(car command)' in this hunk:

> +          ;; Reset the prompt if the command we just aborted was in the
> +          ;; foreground.
> +          (unless (car command)
> +            (declare-function eshell-reset "esh-mode" (&optional no-hooks))
> +            (eshell-reset)))))))

IOW, why '(car command)' is used as an indication of a fore/background
command?  Also, why does the comment say "foreground" while your text
says we don't want the prompt if the killed program was in the
background?

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Jim Porter]

On 7/10/2024 4:16 AM, Eli Zaretskii wrote:
>> Date: Tue, 9 Jul 2024 11:04:05 -0700
>> From: Jim Porter <jporterbugs@gmail.com>
>>
>> This is a regression from Emacs 29, likely due to some changes I made to
>> improve support for complex background commands. Eli, is this ok to
>> merge to the release branch?
> 
> I don't think I understand the essence of the change, and thus cannot
> appreciate its effects enough to be able to answer this.  What is the
> significance of '(car command)' in this hunk:

'command' is a "command entry", and the result of 
'eshell-commands-for-process', which returns a list of elements of the form:

     (BACKGROUND FORM PROCESSES)

BACKGROUND is non-nil if the command is being run in the background.

>> +          ;; Reset the prompt if the command we just aborted was in the
>> +          ;; foreground.
>> +          (unless (car command)
>> +            (declare-function eshell-reset "esh-mode" (&optional no-hooks))
>> +            (eshell-reset)))))))
> 
> IOW, why '(car command)' is used as an indication of a fore/background
> command?  Also, why does the comment say "foreground" while your text
> says we don't want the prompt if the killed program was in the
> background?

We want to reset the prompt (this just emits a new command prompt) for 
foreground commands, but for background commands, we don't need to do 
anything. Would it be clearer if I inverted the wording in the comment, 
like, "Don't reset the prompt if the command we just aborted was in the 
background"?

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Eli Zaretskii]

> Date: Wed, 10 Jul 2024 09:16:11 -0700
> Cc: 72018@debbugs.gnu.org
> From: Jim Porter <jporterbugs@gmail.com>
> 
> On 7/10/2024 4:16 AM, Eli Zaretskii wrote:
> > 
> > I don't think I understand the essence of the change, and thus cannot
> > appreciate its effects enough to be able to answer this.  What is the
> > significance of '(car command)' in this hunk:
> 
> 'command' is a "command entry", and the result of 
> 'eshell-commands-for-process', which returns a list of elements of the form:
> 
>      (BACKGROUND FORM PROCESSES)
> 
> BACKGROUND is non-nil if the command is being run in the background.
> 
> >> +          ;; Reset the prompt if the command we just aborted was in the
> >> +          ;; foreground.
> >> +          (unless (car command)
> >> +            (declare-function eshell-reset "esh-mode" (&optional no-hooks))
> >> +            (eshell-reset)))))))
> > 
> > IOW, why '(car command)' is used as an indication of a fore/background
> > command?  Also, why does the comment say "foreground" while your text
> > says we don't want the prompt if the killed program was in the
> > background?
> 
> We want to reset the prompt (this just emits a new command prompt) for 
> foreground commands, but for background commands, we don't need to do 
> anything. Would it be clearer if I inverted the wording in the comment, 
> like, "Don't reset the prompt if the command we just aborted was in the 
> background"?

I think these subtleties just warrant more detailed comments, and then
we'll be fine.

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Jim Porter]

[-- Attachment #1: Type: text/plain, Size: 583 bytes --]

On 7/10/2024 10:34 AM, Eli Zaretskii wrote:
>> Date: Wed, 10 Jul 2024 09:16:11 -0700
>> Cc: 72018@debbugs.gnu.org
>> From: Jim Porter <jporterbugs@gmail.com>
>>
>> We want to reset the prompt (this just emits a new command prompt) for
>> foreground commands, but for background commands, we don't need to do
>> anything. Would it be clearer if I inverted the wording in the comment,
>> like, "Don't reset the prompt if the command we just aborted was in the
>> background"?
> 
> I think these subtleties just warrant more detailed comments, and then
> we'll be fine.

How about this?

[-- Attachment #2: 0001-Don-t-emit-a-prompt-in-Eshell-when-a-background-comm.patch --]
[-- Type: text/plain, Size: 3394 bytes --]

From 84fb997a94503e76a65a5b96301d9ff2baac1fc3 Mon Sep 17 00:00:00 2001
From: Jim Porter <jporterbugs@gmail.com>
Date: Tue, 9 Jul 2024 10:45:35 -0700
Subject: [PATCH] Don't emit a prompt in Eshell when a background command is
 killed

* lisp/eshell/esh-cmd.el (eshell-resume-command): Check for
background-ness before resetting the prompt.

* test/lisp/eshell/esh-cmd-tests.el
(esh-cmd-test/background/simple-command): Make the regexp a bit
stricter.
(esh-cmd-test/background/kill): New test.
---
 lisp/eshell/esh-cmd.el            | 11 +++++++++--
 test/lisp/eshell/esh-cmd-tests.el | 14 +++++++++++++-
 2 files changed, 22 insertions(+), 3 deletions(-)

diff --git a/lisp/eshell/esh-cmd.el b/lisp/eshell/esh-cmd.el
index 0b3137127d2..e97e4f6d067 100644
--- a/lisp/eshell/esh-cmd.el
+++ b/lisp/eshell/esh-cmd.el
@@ -1030,6 +1030,9 @@ eshell-resume-command
 PROC is the process that invoked this from its sentinel, and
 STATUS is its status."
   (when proc
+    ;; Iterate over all the commands associated with this process.  Each
+    ;; element is a list of the form (BACKGROUND FORM PROCESSES) (see
+    ;; `eshell-add-command').
     (dolist (command (eshell-commands-for-process proc))
       (unless (seq-some #'eshell-process-active-p (nth 2 command))
         (setf (nth 2 command) nil) ; Clear processes from command.
@@ -1040,8 +1043,12 @@ eshell-resume-command
                  (not (string-match eshell-reset-signals status)))
             (eshell-resume-eval command)
           (eshell-remove-command command)
-          (declare-function eshell-reset "esh-mode" (&optional no-hooks))
-          (eshell-reset))))))
+          ;; Check if the command we just aborted is marked as a
+          ;; background command.  If not, we need to reset the prompt so
+          ;; the user can enter another command.
+          (unless (car command)
+            (declare-function eshell-reset "esh-mode" (&optional no-hooks))
+            (eshell-reset)))))))
 
 (defun eshell-resume-eval (command)
   "Destructively evaluate a COMMAND which may need to be deferred.
diff --git a/test/lisp/eshell/esh-cmd-tests.el b/test/lisp/eshell/esh-cmd-tests.el
index 70e1901c169..d8124a19af6 100644
--- a/test/lisp/eshell/esh-cmd-tests.el
+++ b/test/lisp/eshell/esh-cmd-tests.el
@@ -113,7 +113,7 @@ esh-cmd-test/background/simple-command
     (with-temp-eshell
      (eshell-match-command-output
       (format "*echo hi > #<%s> &" bufname)
-      (rx "[echo" (? ".exe") "] " (+ digit) "\n"))
+      (rx bos "[echo" (? ".exe") "] " (+ digit) "\n"))
      (eshell-wait-for-subprocess t))
     (should (equal (buffer-string) "hi\n"))))
 
@@ -129,6 +129,18 @@ esh-cmd-test/background/subcommand
      (eshell-wait-for-subprocess t))
     (should (equal (buffer-string) "olleh\n"))))
 
+(ert-deftest esh-cmd-test/background/kill ()
+  "Make sure that a background command that gets killed doesn't emit a prompt."
+  (skip-unless (executable-find "sleep"))
+  (let ((background-message (rx bos "[sleep" (? ".exe") "] " (+ digit) "\n")))
+    (with-temp-eshell
+      (eshell-match-command-output "*sleep 10 &" background-message)
+      (kill-process (caar eshell-process-list))
+      (eshell-wait-for-subprocess t)
+      ;; Ensure we didn't emit another prompt after killing the
+      ;; background process.
+      (should (eshell-match-output background-message)))))
+
 \f
 ;; Lisp forms
 
-- 
2.25.1

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Eli Zaretskii]

> Date: Wed, 10 Jul 2024 12:55:25 -0700
> Cc: 72018@debbugs.gnu.org
> From: Jim Porter <jporterbugs@gmail.com>
> 
> On 7/10/2024 10:34 AM, Eli Zaretskii wrote:
> >> Date: Wed, 10 Jul 2024 09:16:11 -0700
> >> Cc: 72018@debbugs.gnu.org
> >> From: Jim Porter <jporterbugs@gmail.com>
> >>
> >> We want to reset the prompt (this just emits a new command prompt) for
> >> foreground commands, but for background commands, we don't need to do
> >> anything. Would it be clearer if I inverted the wording in the comment,
> >> like, "Don't reset the prompt if the command we just aborted was in the
> >> background"?
> > 
> > I think these subtleties just warrant more detailed comments, and then
> > we'll be fine.
> 
> How about this?

LGTM, thanks.

This is okay for emacs-30.

bug#72018: 30.0.60; [PATCH] Don't emit a prompt when a background Eshell process is killed

[Jim Porter]

On 7/10/2024 9:34 PM, Eli Zaretskii wrote:
> LGTM, thanks.
> 
> This is okay for emacs-30.

Thanks. Pushed to the release branch as 0de0056fd6b, and closing this bug.