[slime-devel] Re: Question about decoupling load-server and start-server in slime-init-command

[Helmut Eller]

* Juan M. Bello Rivas [2007-11-16 15:41+0100] writes:

> Hi,
>
> I have a slow system where launching slime (current CVS version) takes
> several seconds.  Wishing to cut down the initialization time, I
> dumped a sbcl core with swank preloaded but slime didn't start faster
> because swank was being reloaded.  To avoid this, I modified
> slime-init-command and now slime starts lightning fast.
>
> There's a comment before slime-init-command saying:
>
> ;; XXX load-server & start-server used to be separated. maybe that was  better.
>
> Why were the two functionalities merged?  

Previously we sent two forms.  Due to bugs, misconfiguration, or
whatever, there are sometimes errors during the first step, but since
the second form was buffered in the input stream it was also executed.

This was confusing--at least I searched the error in the wrong place
several times--because you would end up in the (nested) debugger
thinking that the second step failed.

This doesn't happen anymore, because we now send everything as a
single form, side-stepping the problem with input-buffering.

> I definitely think it was better.  

I don't see a clear advantage, because if you use a non-standard
startup procedure you need to configure something anyway and creating
the start-server part isn't a big deal.

> Perhaps I missed an obvious way to achieve the same goal?

The section "Loading Swank faster" of the manual describes a way to
achieve it without modifying slime-init-command directly.

Helmut.