Re: `scm_c_read ()' and `swap_buffer' trick harmful

[ludo@gnu.org (Ludovic Courtès)]

Hello,

Miroslav Lichvar <mlichvar@redhat.com> writes:

> On Sun, Nov 23, 2008 at 10:30:19PM +0000, Neil Jerram wrote:
>> 2008/11/21 Ludovic Courtès <ludo@gnu.org>:
>> > I confirm it works with CBIPs in Guile-R6RS-Libs, but I had to make this
>> > small change:
>> >
>> >  http://repo.or.cz/w/guile-r6rs-libs.git?a=commitdiff;h=dfab2fe21f82bc40cfdf9a3a3eeb6936d5935c8c
>> >
>> > IOW, with Guile up to 1.8.5, it was harmless to leave `read_buf_size'
>> > uninitialized (actually, zeroed), which was OK since CBIPs don't use it,
>> > but the change in `scm_c_read ()' makes it necessary to initialize it.
>> > It's probably a reasonable expectation, though.
>> 
>> Hmm.  I agree that this isn't ideal.  I guess it's OK for now though.
>> 
>> > Please commit!
>> 
>> Thanks, doing that now.
>
> It seems that the optimizations broke lilypond, it's now failing with
> this message:
>
> ports.c:978: scm_fill_input: Assertion `pt->read_pos == pt->read_end' failed.

Ouch!

> A bug report is here:
> https://bugzilla.redhat.com/show_bug.cgi?id=477007

Can you get a full C backtrace?

A quick grep reveals that Lilypond does not create any custom port type,
which isn't reassuring...

Thanks,
Ludo'.