Re: Paralellizing Lilypond [was: Re: Sibelius Software UK office shutsdown]

[Han-Wen Nienhuys]

On Sat, Aug 11, 2012 at 4:18 PM, David Kastrup <address@hidden> wrote:

>>>> I believe it would be a compilable file.
>>>
>>>
>>> Useful to know, thank you!
>>>
>>> ... but I think it emphasizes my real point: this puts the onus on the user
>>> to split up a project into independently-compilable units.  I think that
>>> it's worth having Lilypond try and automatically identify independent units,
>>> which could have knock-on benefits in terms of minimizing rebuild times for
>>> scores.
>>
>> It would be nice if this were automatically splittable, but the
>> reality is that GUILE has no meaningful multithreading support at the
>> interpreter level, so almost mutating operation has the potential to
>> be a race condition.
>
> Guile 2.0 has threading support.  Whether that will prove usefully
> applicable to LilyPond will be a different question.

It has had threading for a long time, but the interpreter is full of
global variables.  This makes it difficult to ensure that operations
are properly serialized.  Also, I would be surprised if the MT has had
a lot of real testing.

--
Han-Wen Nienhuys - address@hidden - http://www.xs4all.nl/~hanwen