Re: Making Lilypond faster

[Thomas Morley]

Am Mi., 3. Juni 2020 um 14:45 Uhr schrieb Paolo Prete <paolopr976@gmail.com>:
>
>
>>
>>
>> Hi Paolo,
>>
>> The suggestion to focus on guile-3 makes no sense at the moment,
>> because it is not broadly in the distros, afaict.
>>
>>
>
>
> Hello Harm,
>
> what about this?
>
> http://lilypond.1069038.n5.nabble.com/guile-3-0-and-LilyPond-here-input-regression-context-defaultchild-cycle-ly-fails-td227492.html#a227495
>
> Did you succeed (or someone else) in compiling with guile-3, meanwhile ?
> If so, did you see some improvement in speed?
>
> Thanks!
> Best,
> P
>
>
>

Hi Paolo,

today I tried to compile LilyPond with guile-3.

####

Guile-3 was compiled from their repo with:

sh autogen.sh
./configure
make
sudo make install

giving me "3.0.2.124-e7398")

####

For compiling LilyPond I used:

sh autogen.sh --noconfigure
mkdir -p build/
cd build/
../configure GUILE_FLAVOR=guile-3.0
make

###

`make' succeeded, though there is a problem reported:
imported module (lily) overrides core binding `_'
Which is caused by guile now using `_' as a reserved keyword. Formerly
it was short for gettext, now `G_' should be used.
Though, for now I didn't fix it.

I then compiled some scores with this LilyPond and compared timing
with released lilypond-2.21.1.:
LilyPond with guile-3 needs in average ~40% more time.
No clue whether the timing is influenced by the not fixed gettext-shortcut.

Note that I didn't try to build the docs nor did I try to compile the regtests.
More problems could pop up then...


Cheers,
  Harm