Re: Problem with guile-2.9.1-prerelease

[Thomas Morley]

Am Fr., 12. Okt. 2018 um 01:45 Uhr schrieb Thomas Morley
<address@hidden>:

> Tomorrow I'll redo a full 'make doc'.
> Testing your changes with guile-2.2.4 and guile-1.8 is postponed for
> tomorrow as well.

To facilitate testing I had to change my local setup to compare
multiple combinations of guile-versions with lilypond.
Sorry for the delay, compiling guile-versions is very time-consuming...

So, here my setup, in the end I tested with 5 lilypond-versions:

(1)
LilyPond 2.19.82 from the installer, i.e. with guile-1.8

Below 4 selfcompiled versions out of

commit ea638182bcc87414c7f186d40f376bbbf560f5d1
Author: David Kastrup <address@hidden>
Date:   Wed Oct 3 14:20:45 2018 +0200

    Issue 5423: First separator for subassignments must be '.'

    This pares down syntax supported since issue 4790 to match the allowed
    usage from issue 4797.  Permitting ',' here seemed particularly
    strange.

(2)
LilyPond 2.21.0 with guile-1.8.8

(3)
LilyPond 2.21.0 with guile-2.0.14

(4)
LilyPond 2.21.0 with guile-2.2.4

(5)
LilyPond 2.21.0 with guile-2.9.1

(3) to (5) have the attached patches applied to make them work with guilev2

on top of (2) - (5) David's patch from this thread:
    Use different `values' implementation of Guilev2
is applied. It is in the attached archive as well.


Results:

David, your patch always works and does as desired.
How about putting it up for review?

Karlin et all, here some performance-values (always taken from a
second of two runs) with
$ time <lilypond-command> <file.ly>

>From a file with close to no user-generated guile-code
(resulting in a 40-pages-pdf)

ad (1)
real    1m19,297s
user    1m16,390s
sys    0m1,883s

ad (2)
real    1m10,707s
user    1m9,220s
sys    0m1,336s


ad (3)
real    4m13,883s
user    5m7,904s
sys    0m1,474s


ad (4)
real    4m3,027s
user    5m10,502s
sys    0m1,697s

ad (5)
real    3m34,525s
user    4m34,974s
sys    0m1,613s


>From a file with huge amount of user generated guile-code
(resulting in a 8-pages-pdf)

(1)
real    0m24,107s
user    0m23,002s
sys    0m1,101s

(2)
real    0m21,689s
user    0m20,740s
sys    0m0,923s

(3)
real    1m20,443s
user    1m33,126s
sys    0m0,918s

(4)
real    0m45,537s
user    0m52,817s
sys    0m0,991s

(5)
real    0m40,445s
user    0m46,441s
sys    0m0,955s

So there _is_ some improvement, but all in all not overwhelming, imho.

Additionally, I've probably found a new small issue with guilev2, but
this is worth another thread.

As said above the used guilev2-patches are attached, if someone wants
to join testing.
Be aware some of them (especially my own ones) are more workarounds
than proper fixes.
Hints are always welcome.

Cheers,
  Harm

patches-for-guile-2-9-1.zip
Description: Zip archive