Re: Problem with guile-2.9.1-prerelease

[David Kastrup]

Carl Sorensen <address@hidden> writes:

> On 10/11/18, 12:35 PM, "lilypond-devel on behalf of Urs Liska"
> <address@hidden on behalf of
> address@hidden> wrote:
>
>     
>     
>     Am 11. Oktober 2018 20:17:45 MESZ schrieb Karlin High <address@hidden>:
>     >On 10/11/2018 12:59 PM, David Kastrup wrote:
>     >What level of optimism is appropriate for that claim?
>     
>     I don't think that tells us very much. From 1.8 to 2 they added
> "significant improvements" that caused massive performance problems
> for our specific use case.  And I woulcn't bet too much money that the
> improvements you mention specifically address these issues ...
>     
>
> On the other hand, the 3.0 test that they report on indicates that the
> new JIT compile code is almost as fast as 1.8.

That makes little to no sense since 1.8 was slower than Guile 2 for most
compiled code.

The only thing I can imagine is that they mean the new JIT compile code
is almost as fast as programming in C while manipulating SCM data with
the 1.8 API calls.

> And the way around the performance issues with 2 was to use
> precompiled bytecode because the compiler was slow.  This new JIT
> compile claims to not need bytecode, so I am optimistic....

Our problems is not code written in Scheme but passing stuff between C++
and Scheme.  Let's hope your optimism turns out more vindicated than my
pessimism.

-- 
David Kastrup