Re: Parsed object should be dead

[Jonas Hahnfeld]

Am Sonntag, dem 07.02.2021 um 23:16 +0100 schrieb Han-Wen Nienhuys:
> On Sat, Feb 6, 2021 at 7:33 PM Dan Eble <dan@faithful.be> wrote:
> 
> > Did something change in the past couple of weeks to make "Parsed object
> > should be dead" messages more likely?  Now that I've turned from
> > documentation back to development, they seem to be getting in my way much
> > more than they used to.
> > 
> > This stands out:
> > 
> > commit d1be5ec60a058e363835bee60ece6fc61a67c5a9
> > Author: Han-Wen Nienhuys <hanwenn@gmail.com>
> > Date:   Sun Jan 24 16:50:19 2021 +0100
> > 
> >     Introduce 'debug-gc-lifetimes option
> > 
> >     The debug-gc-assert-parsed-dead feature has been documented, but it is
> >     really an internal mechanism (it only has effect during garbage
> >     collection, and has to be coupled with calls to ly:parsed-undead-list!
> >     in order to do something useful.
> > 
> >     Instead, introduce the debug-gc-object-lifetimes option, which
> >     controls the entire check (ie. the extra GC call, and the printing of
> >     objects afterwards). For backward compatibility, this feature defaults
> >     to #t.
> > 
> > I wonder if we could disable this by default.  As a user, I don't like
> > getting 672 lines of console output when I would have gotten 31 with this
> > option disabled.
> > 
> 
> I think we should default debug-gc-object-lifetimes to #f, and set it to #t
> for the regression test suite.

Note that --enable-checking is off by default already, so the check
actually doesn't do anything. Also, as far as I understand, once we
disable 'debug-gc-object-lifetimes there will be no (gc) between the
files and the GC bug in the released versions of Guile 1.8 will
trigger, leading to excessive memory usage.

signature.asc
Description: This is a digitally signed message part