Constructor created by define-configuration is slow (>25s) when many fields are defined

Open

Details

2 participants

Ludovic Courtès
Tomas Volf

Owner: unassigned

Submitted by: Tomas Volf

Severity: normal

Tomas Volf wrote on 9 Dec 2024 16:13

Recipients:(address . bug-guix@gnu.org)

Message-ID:87msh499f7.fsf@wolfsden.cz

Hello,

I am writing a configuration for mpv and hit an issue of the generated

constructor (`home-mpv-configuration') is extremely slow. While the

almost 800 fields is unusually high, initializing the instances of the

record should *not* take almost 30 seconds. The full file is attached

to this email (as `mpv.scm').

Toggle snippet (6 lines)

scheme@(guix-user)> ,use (gnu home services mpv)

scheme@(guix-user)> ,t (and (home-mpv-configuration (global (mpv-profile-configuration (fullscreen #t)))) #t)

$1 = #t

;; 29.615315s real time, 50.896973s run time. 24.031723s spent in GC.

Interestingly, profiling does not give us anything useful:

Toggle snippet (4 lines)

scheme@(guix-user)> ,pr (and (home-mpv-configuration (global (mpv-profile-configuration (fullscreen #t)))) #t)

No samples recorded.

I also tried to remove all my custom sanitizers, to make sure the issue

is not in my code, but that did not help.

I took a short look, but code of both the define-configuration and of

Guix records is complex, full of syntax magic and I did not have any

luck trying to debug it myself. Hence this report.

Have a nice day,

Tomas

Attachment: mpv.scm

-- 
There are only two hard things in Computer Science:
cache invalidation, naming things and off-by-one errors.

Ludovic Courtès wrote on 24 Dec 2024 15:38

Recipients:(name . Tomas Volf)(address . ~@wolfsden.cz)(address . 74748@debbugs.gnu.org)

Message-ID:87cyhh5eop.fsf@gnu.org

Hi,

Tomas Volf <~@wolfsden.cz> skribis:

Toggle quote (6 lines)

> I am writing a configuration for mpv and hit an issue of the generated

> constructor (`home-mpv-configuration') is extremely slow. While the

> almost 800 fields is unusually high, initializing the instances of the

> record should *not* take almost 30 seconds. The full file is attached

> to this email (as `mpv.scm').

AFAICS, time is spent in one of the compilation steps:

Toggle snippet (13 lines)scheme@(gnu home services mpv)> ,o interp #t
scheme@(gnu home services mpv)> ,t (lambda () (mpv-profile-configuration))
$13 = #<procedure 7f6ed3796700 at ice-9/eval.scm:330:13 ()>
;; 0.376347s real time, 0.375402s run time.  0.000000s spent in GC.
scheme@(gnu home services mpv)> ,o interp #f
scheme@(gnu home services mpv)> ,t (lambda () (mpv-profile-configuration))
$14 = #<procedure 7f6ec32088f8 at <unknown port>:40:3 ()>
;; 11.149828s real time, 12.052915s run time.  1.581736s spent in GC.
scheme@(gnu home services mpv)> ,t (->bool (macroexpand '(mpv-profile-configuration)))
$15 = #t
;; 0.373865s real time, 0.372698s run time.  0.000000s spent in GC.

If we look more closely, it’s the optimizing compiler that takes time;

the baseline compile (-O1) is doing okay:

Toggle snippet (16 lines)scheme@(gnu home services mpv)> ,use(system base compile)
scheme@(gnu home services mpv)> ,t (->bool (compile '(mpv-profile-configuration) #:to 'tree-il #:env (current-module)))
$20 = #t
;; 0.378741s real time, 0.377043s run time.  0.000000s spent in GC.
scheme@(gnu home services mpv)> ,t (->bool (compile '(mpv-profile-configuration) #:to 'bytecode #:env (current-module)))
$21 = #t
;; 11.946879s real time, 12.961704s run time.  1.777478s spent in GC.
scheme@(gnu home services mpv)> (define til (compile '(mpv-profile-configuration) #:to 'tree-il #:env (current-module)))
scheme@(gnu home services mpv)> ,t (->bool (compile til #:from 'tree-il #:to 'bytecode))
$22 = #t
;; 11.403420s real time, 12.317183s run time.  1.581972s spent in GC.
scheme@(gnu home services mpv)> ,t (->bool (compile til #:from 'tree-il #:to 'bytecode #:optimization-level 1))
$23 = #t
;; 0.225455s real time, 0.156137s run time.  0.000000s spent in GC.

Currently (guix build compile) applies these options to services:

((string-contains file "gnu/services/")

;; '-O2 -Ono-letrectify' compiles about ~20% faster than '-O2' for

;; large files like gnu/services/mail.scm.

(override-option #:letrectify? #f

(optimizations-for-level 2)))

I think we should at least apply this patch:

Toggle diff (14 lines)diff --git a/guix/build/compile.scm b/guix/build/compile.scm
index 5b27b55d02..f90016b9ae 100644
--- a/guix/build/compile.scm
+++ b/guix/build/compile.scm
@@ -127,7 +127,8 @@ (define (optimization-options file)
          ;; Use '-O1' to have partial evaluation and primitive inlining so we
          ;; can honor the "macro writer's bill of rights".
          (optimizations-for-level 1))
-        ((string-contains file "gnu/services/")
+        ((or (string-contains file "gnu/services/")
+             (string-contains file "gnu/home/services/"))
          ;; '-O2 -Ono-letrectify' compiles about ~20% faster than '-O2' for
          ;; large files like gnu/services/mail.scm.
          (override-option #:letrectify? #f

Would that be enough for the home-mpv configuration records you wrote?

Thanks,

Ludo’.

Tomas Volf wrote on 27 Dec 2024 23:50

Recipients:(name . Ludovic Courtès)(address . ludo@gnu.org)(address . 74748@debbugs.gnu.org)

Message-ID:87a5cg4u75.fsf@wolfsden.cz

Hello,

thank you for finding the time to look into this issue. Replies below.

Ludovic Courtès <ludo@gnu.org> writes:

Toggle quote (71 lines)> Hi,
>
> Tomas Volf <~@wolfsden.cz> skribis:
>
>> I am writing a configuration for mpv and hit an issue of the generated
>> constructor (`home-mpv-configuration') is extremely slow.  While the
>> almost 800 fields is unusually high, initializing the instances of the
>> record should *not* take almost 30 seconds.  The full file is attached
>> to this email (as `mpv.scm').
>
> AFAICS, time is spent in one of the compilation steps:
>
> scheme@(gnu home services mpv)> ,o interp #t
> scheme@(gnu home services mpv)> ,t (lambda () (mpv-profile-configuration))
> $13 = #<procedure 7f6ed3796700 at ice-9/eval.scm:330:13 ()>
> ;; 0.376347s real time, 0.375402s run time.  0.000000s spent in GC.
> scheme@(gnu home services mpv)> ,o interp #f
> scheme@(gnu home services mpv)> ,t (lambda () (mpv-profile-configuration))
> $14 = #<procedure 7f6ec32088f8 at <unknown port>:40:3 ()>
> ;; 11.149828s real time, 12.052915s run time.  1.581736s spent in GC.
> scheme@(gnu home services mpv)> ,t (->bool (macroexpand '(mpv-profile-configuration)))
> $15 = #t
> ;; 0.373865s real time, 0.372698s run time.  0.000000s spent in GC.
>
>
> If we look more closely, it’s the optimizing compiler that takes time;
> the baseline compile (-O1) is doing okay:
>
> scheme@(gnu home services mpv)> ,use(system base compile)
> scheme@(gnu home services mpv)> ,t (->bool (compile '(mpv-profile-configuration) #:to 'tree-il #:env (current-module)))
> $20 = #t
> ;; 0.378741s real time, 0.377043s run time.  0.000000s spent in GC.
> scheme@(gnu home services mpv)> ,t (->bool (compile '(mpv-profile-configuration) #:to 'bytecode #:env (current-module)))
> $21 = #t
> ;; 11.946879s real time, 12.961704s run time.  1.777478s spent in GC.
> scheme@(gnu home services mpv)> (define til (compile '(mpv-profile-configuration) #:to 'tree-il #:env (current-module)))
> scheme@(gnu home services mpv)> ,t (->bool (compile til #:from 'tree-il #:to 'bytecode))
> $22 = #t
> ;; 11.403420s real time, 12.317183s run time.  1.581972s spent in GC.
> scheme@(gnu home services mpv)> ,t (->bool (compile til #:from 'tree-il #:to 'bytecode #:optimization-level 1))
> $23 = #t
> ;; 0.225455s real time, 0.156137s run time.  0.000000s spent in GC.
>
> Currently (guix build compile) applies these options to services:
>
>         ((string-contains file "gnu/services/")
>          ;; '-O2 -Ono-letrectify' compiles about ~20% faster than '-O2' for
>          ;; large files like gnu/services/mail.scm.
>          (override-option #:letrectify? #f
>                           (optimizations-for-level 2)))
>
> I think we should at least apply this patch:
>
> diff --git a/guix/build/compile.scm b/guix/build/compile.scm
> index 5b27b55d02..f90016b9ae 100644
> --- a/guix/build/compile.scm
> +++ b/guix/build/compile.scm
> @@ -127,7 +127,8 @@ (define (optimization-options file)
>           ;; Use '-O1' to have partial evaluation and primitive inlining so we
>           ;; can honor the "macro writer's bill of rights".
>           (optimizations-for-level 1))
> -        ((string-contains file "gnu/services/")
> +        ((or (string-contains file "gnu/services/")
> +             (string-contains file "gnu/home/services/"))
>           ;; '-O2 -Ono-letrectify' compiles about ~20% faster than '-O2' for
>           ;; large files like gnu/services/mail.scm.
>           (override-option #:letrectify? #f
>
>
> Would that be enough for the home-mpv configuration records you wrote?

I admit I did not test the patch attached, but my understanding is that
even if the speed up was sufficient, it would take effect only for
compilation via the make command.  Assuming that is correct, it would
not change much.  I am (luckily) former C++ developer, so I am used to
compilations taking ~forever.  So, yes, while the compilation of mpv.scm
(via `make') took roughly 40s on my machine, it was annoying, but not
worthy of a bug report.

The actual problem for me was the slowness in REPL.  Would that be
addressed by the patch above?  If not, do we have any way to do that?

)tomas

-- 
There are only two hard things in Computer Science:
cache invalidation, naming things and off-by-one errors.

Your comment

Commenting via the web interface is currently disabled.

To comment on this conversation send an email to 74748@debbugs.gnu.org

To respond to this issue using the mumi CLI, first switch to it

mumi current 74748

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

is:open	open issues
is:done	closed issues
submitter:<who>	search issue submitter
author:<who>	search by message author
date:yesterday..now	search by issue date
mdate:3m..2d	search by message date