[Installer] ‘guix system init’ displays dots instead of progress bars

  • Done
  • quality assurance status badge
Details
2 participants
  • Ludovic Courtès
  • Mathieu Othacehe
Owner
unassigned
Submitted by
Ludovic Courtès
Severity
important
Merged with
L
L
Ludovic Courtès wrote on 11 May 2022 11:23
[Installer] ‘guix system init ’ displays dots instead of progress bars
(address . bug-guix@gnu.org)
87pmkko21t.fsf@gnu.org
Hi!

I’ve used an installer built from this commit:

Toggle snippet (7 lines)
Generation 214 May 02 2022 21:44:14 (current)
guix 6b588da
repository URL: https://git.savannah.gnu.org/git/guix.git
branch: master
commit: 6b588da368c77cde82ea2f22ca315116228777ad

During installation, ‘guix system init’ displays dots instead of
progress bars for downloads and such.

This suggests that ‘display-download-progress’ gets #:tty? #f. This may
be a side effect of running it in a pipe in
‘run-external-command-with-handler’.

Thoughts?

Ludo’.
L
L
Ludovic Courtès wrote on 11 May 2022 11:26
control message for bug #55360
(address . control@debbugs.gnu.org)
87o804o1wl.fsf@gnu.org
severity 55360 important
quit
L
L
Ludovic Courtès wrote on 11 May 2022 11:27
control message for bug #53214
(address . control@debbugs.gnu.org)
87mtfoo1w1.fsf@gnu.org
block 53214 by 55360
quit
L
L
Ludovic Courtès wrote on 18 Jul 2022 21:58
Re: bug#55360: [Installer] ‘guix system init’ displays dots instead of progress bars
(address . 55360@debbugs.gnu.org)
87y1wqw5tx.fsf@gnu.org
Hi!

Ludovic Courtès <ludo@gnu.org> skribis:

Toggle quote (7 lines)
> During installation, ‘guix system init’ displays dots instead of
> progress bars for downloads and such.
>
> This suggests that ‘display-download-progress’ gets #:tty? #f. This may
> be a side effect of running it in a pipe in
> ‘run-external-command-with-handler’.

This was introduced by:

commit 0b9fbbb4dd24f227c9a708561ba291f6169ad2e6
Date: Sat Jan 15 14:50:00 2022 +0100

installer: Capture external commands output.

* gnu/installer/utils.scm (run-external-command-with-handler,
run-external-command-with-line-hooks): New variables.
(run-command): Use run-external-command-with-line-hooks.

(unsurprisingly) to log standard output of commands.

I’m not sure it’s a good idea for ‘guix system init’: we’d be logging
mostly progress bars, package names, and the likes to syslog—not super
useful. So I’d suggest not capturing stdout of ‘guix system init’.

However, I wouldn’t mind factorizing the syslog facilities from
2cf65e1d543407bc7db43e7c7d39a215907efebc into, say, (guix syslog), and
have (guix scripts system) use it to log important events.

Thoughts?

What about using ‘run-external-command-with-line-hooks’ for other
commands though? IIUC, the primary use case is file system tools.

My gut feeling is that the cost/benefit ratio isn’t worth it (the cost
here is complexity and maintenance burden of
‘run-external-command-with-handler’ & co.; the benefit is improved
logging).

WDYT?

I can hopefully dedicate a bit of time to this, but I need your
guidance, comrades. :-)

Ludo’.
L
L
Ludovic Courtès wrote on 9 Oct 2022 18:56
control message for bug #55360
(address . control@debbugs.gnu.org)
878rlp5418.fsf@gnu.org
merge 55360 58375
quit
M
M
Mathieu Othacehe wrote on 13 Oct 2022 09:04
Re: bug#58375: Installer does not show what is being downloaded
(name . Ludovic Courtès)(address . ludo@gnu.org)
87lepk5hnh.fsf_-_@gnu.org
Hey Ludo!

Toggle quote (4 lines)
> I’m not sure it’s a good idea for ‘guix system init’: we’d be logging
> mostly progress bars, package names, and the likes to syslog—not super
> useful. So I’d suggest not capturing stdout of ‘guix system init’.

In the bug report https://issues.guix.gnu.org/57983capturing the 'guix
system init' output highlighted a "guix substitute" crash. So it does
seem like a useful mechanism, especially while 56005 is still open.

Now the current situation is also not really acceptable. What about
hiding the "guix system init" output completely and display a progress
bar page instead?

Thanks,

Mathieu
L
L
Ludovic Courtès wrote on 13 Oct 2022 15:21
(name . Mathieu Othacehe)(address . othacehe@gnu.org)
87zgdzq2oe.fsf@gnu.org
Hi!

Mathieu Othacehe <othacehe@gnu.org> skribis:

Toggle quote (8 lines)
>> I’m not sure it’s a good idea for ‘guix system init’: we’d be logging
>> mostly progress bars, package names, and the likes to syslog—not super
>> useful. So I’d suggest not capturing stdout of ‘guix system init’.
>
> In the bug report https://issues.guix.gnu.org/57983 capturing the 'guix
> system init' output highlighted a "guix substitute" crash. So it does
> seem like a useful mechanism, especially while 56005 is still open.

installer-specific, so it’s annoying that we have to prepare for that,
but we can’t deny such bugs exist and prevent installation.

If we really want to capture the output of ‘guix system init’, then we
need to open a pseudo-terminal with ‘openpty’ & co. instead of ‘pipe’ in
‘run-external-command-with-handler’. That may be relatively easy
actually.

But then I think this should be used sparsely, maybe only for ‘guix
system init’.

Toggle quote (2 lines)
> Now the current situation is also not really acceptable.

Nope. :-)

Toggle quote (3 lines)
> What about hiding the "guix system init" output completely and display
> a progress bar page instead?

I don’t think we can do that (with grafts, only part of the build plan
is known upfront so we don’t even know beforehand how many items will be
built/downloaded).

Also, I think there are two strategies: either we run ‘guix system
init’, in which case we let its output through, or we integrate ‘guix
system init’ functionality in the installer so we have more fine-grain
control over the process, in which case we could also have more
graphical output or something.

That second solution is a lot of work, though.

Thanks,
Ludo’.
M
M
Mathieu Othacehe wrote on 14 Oct 2022 17:44
(name . Ludovic Courtès)(address . ludo@gnu.org)
871qrabeav.fsf_-_@gnu.org
Hey,

Toggle quote (5 lines)
> If we really want to capture the output of ‘guix system init’, then we
> need to open a pseudo-terminal with ‘openpty’ & co. instead of ‘pipe’ in
> ‘run-external-command-with-handler’. That may be relatively easy
> actually.

So I implemented your proposal. It seems to be working quite well. As
discussed on #guix, we could avoid to dump the download bars to the
syslog if the "guix system init" command succeeds. However, it seems
quite tricky in the current implementation where the syslog dumping is
actually a hook (%syslog-line-hook).

Fixing this issue, I also realized that when the "guix system init"
command fails, the user is only offered to resume the installation or
restart it.

In cases where "guix system init" failed because of a network issue, or
because a partition was too small, restarting/resuming seems like the
right thing to do.

However, when the installer failed because "guix system init" crashed or
segfaulted, restarting/resuming won't probably help, and dumping the
crash is probably the best way to get help. That's why I added in a
second patch, a new button "Report the failure" to the
"run-install-failed-page".

Thanks,

Mathieu
Attachment: 0001-installer-Run-the-guix-system-init-command-in-a-PTY.patch
From 159b82a013cad8c3c698031cce9ee07956153da3 Mon Sep 17 00:00:00 2001
From: Mathieu Othacehe <othacehe@gnu.org>
Date: Fri, 14 Oct 2022 17:33:28 +0200
Subject: [PATCH 2/2] installer: Propose the user to report a "guix system
init" failure.

* gnu/installer/newt/final.scm (run-install-failed-page): Add a "Report the
failure" button.
---
gnu/installer/newt/final.scm | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)

Toggle diff (30 lines)
diff --git a/gnu/installer/newt/final.scm b/gnu/installer/newt/final.scm
index 7c3f73ee82..6e55be5067 100644
--- a/gnu/installer/newt/final.scm
+++ b/gnu/installer/newt/final.scm
@@ -80,16 +80,21 @@ (define (run-install-success-page)
(define (run-install-failed-page)
(match (current-clients)
(()
- (match (choice-window
+ (match (ternary-window
(G_ "Installation failed")
(G_ "Resume")
(G_ "Restart the installer")
+ (G_ "Report the failure")
(G_ "The final system installation step failed. You can resume from \
a specific step, or restart the installer."))
(1 (abort-to-prompt 'installer-step 'abort))
(2
;; Keep going, the installer will be restarted later on.
- #t)))
+ #t)
+ (3 (raise
+ (condition
+ (&message
+ (message "User abort.")))))))
(_
(send-to-clients '(installation-failure))
#t)))
--
2.37.3
L
L
Ludovic Courtès wrote on 18 Oct 2022 17:17
(name . Mathieu Othacehe)(address . othacehe@gnu.org)
87czapma90.fsf@gnu.org
Hi Mathieu,

Mathieu Othacehe <othacehe@gnu.org> skribis:

Toggle quote (6 lines)
> So I implemented your proposal. It seems to be working quite well. As
> discussed on #guix, we could avoid to dump the download bars to the
> syslog if the "guix system init" command succeeds. However, it seems
> quite tricky in the current implementation where the syslog dumping is
> actually a hook (%syslog-line-hook).

Yes.

I haven’t actually tested the patch but it LGTM. One thing to check is
whether ‘terminal-window-size’ returns something sensible for the
pseudo-terminal; it could be that we need an extra ioctl so the
pseudo-terminal has the same size as the actual terminal.

Toggle quote (14 lines)
> Fixing this issue, I also realized that when the "guix system init"
> command fails, the user is only offered to resume the installation or
> restart it.
>
> In cases where "guix system init" failed because of a network issue, or
> because a partition was too small, restarting/resuming seems like the
> right thing to do.
>
> However, when the installer failed because "guix system init" crashed or
> segfaulted, restarting/resuming won't probably help, and dumping the
> crash is probably the best way to get help. That's why I added in a
> second patch, a new button "Report the failure" to the
> "run-install-failed-page".

Neat!

Toggle quote (13 lines)
> From c6286404e9c4c0dc302c3d398a8f27b050cf4ce0 Mon Sep 17 00:00:00 2001
> From: Mathieu Othacehe <othacehe@gnu.org>
> Date: Fri, 14 Oct 2022 17:28:27 +0200
> Subject: [PATCH 1/2] installer: Run the "guix system init" command in a PTY.
>
> Fixes: <https://issues.guix.gnu.org/55360>
>
> * gnu/installer/utils.scm (run-external-command-with-handler/tty): New
> procedure.
> (run-external-command-with-line-hooks, run-command): Add a TTY? argument.
> * gnu/installer/final.scm (install-system): Call run-command with TTY?
> argument set to #true.

LGTM, modulo the terminal size issue mentioned above.

Toggle quote (9 lines)
> From 159b82a013cad8c3c698031cce9ee07956153da3 Mon Sep 17 00:00:00 2001
> From: Mathieu Othacehe <othacehe@gnu.org>
> Date: Fri, 14 Oct 2022 17:33:28 +0200
> Subject: [PATCH 2/2] installer: Propose the user to report a "guix system
> init" failure.
>
> * gnu/installer/newt/final.scm (run-install-failed-page): Add a "Report the
> failure" button.

LGTM.

Thank you!

Ludo’.
M
M
Mathieu Othacehe wrote on 20 Oct 2022 16:40
(name . Ludovic Courtès)(address . ludo@gnu.org)
875ygepnhz.fsf@gnu.org
Hey,

Thanks for having a look!

Toggle quote (5 lines)
> I haven’t actually tested the patch but it LGTM. One thing to check is
> whether ‘terminal-window-size’ returns something sensible for the
> pseudo-terminal; it could be that we need an extra ioctl so the
> pseudo-terminal has the same size as the actual terminal.

Well it returns 0 for all fields, but I tested on several screen sizes
and everything seems fine so I went ahead.

While testing I noticed two new issues though:

1. When the disk is GPT partitionned there is no confirmation page in
"run-label-page". Something I missed in #57232.

2. When there is an exception in run-external-command-with-handler/tty
for instance, the backtrace page is displayed in the PTY and the
keyboard shortcuts do not work anymore.

I'll address point 1 shortly but could use some advice for point 2.

Thanks,

Mathieu
Closed
?
Your comment

This issue is archived.

To comment on this conversation send an email to 55360@debbugs.gnu.org

To respond to this issue using the mumi CLI, first switch to it
mumi current 55360
Then, you may apply the latest patchset in this issue (with sign off)
mumi am -- -s
Or, compose a reply to this issue
mumi compose
Or, send patches to this issue
mumi send-email *.patch