[Uim] uim 1.5.0 TODO list

YAMAMOTO Kengo / YamaKen yamaken at bp.iij4u.or.jp
Thu Jan 25 14:07:46 EET 2007


Hi Etsushi,

At Thu, 25 Jan 2007 17:55:20 +0900,
ek.kato at gmail.com wrote:
> 
> I still don't understand the cause of the bug, using uim_scm_eval()
> for filter_key() in uim-key.c seems to be a workaround.  Does it mean
> uim_scm_call() has a GC related problem?

The bug came from the API design error of
uim_scm_call[1-3](). On a procedure call via these functions,
the evaluated arguments are not GC protected although each
argument evaluations itself are protected.

I'll withdraw the uim_scm_call[0-3]() APIs, and add a set of
replacement for it. Please wait for a while.

At Thu, 25 Jan 2007 15:48:32 +0900,
ek.kato at gmail.com wrote:
> 
> Experimentally, I modified libgcroots's Makefile not to define
> HAVE_GETCONTEXT on Fedora and use __builtin_unwind_init() as in
> original GC6.7 in mach_dep.c, and I've been free from the GC bug so
> far.

libgcroots is not based on GC 6.7 but GC 7.0alpha7. Keep it as
in GC 7.0alpha7.

GC 7 prefers getcontext() over __builtin_unwind_init() if
defined(UNIX_LIKE), regardless of existence of the
HAVE_GETCONTEXT I added in libgcroots. So undefining
HAVE_GETCONTEXT does not affect the context saving method
selection.

------------------------------------------------
YAMAMOTO Kengo / YamaKen  yamaken at bp.iij4u.or.jp
FAMILY   Given / Nick
http://en.wikipedia.org/wiki/Japanese_name



More information about the uim mailing list